1 /*
2 * iterator/iterator.c - iterative resolver DNS query response module
3 *
4 * Copyright (c) 2007, NLnet Labs. All rights reserved.
5 *
6 * This software is open source.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 *
12 * Redistributions of source code must retain the above copyright notice,
13 * this list of conditions and the following disclaimer.
14 *
15 * Redistributions in binary form must reproduce the above copyright notice,
16 * this list of conditions and the following disclaimer in the documentation
17 * and/or other materials provided with the distribution.
18 *
19 * Neither the name of the NLNET LABS nor the names of its contributors may
20 * be used to endorse or promote products derived from this software without
21 * specific prior written permission.
22 *
23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
26 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
27 * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
28 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
29 * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
30 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 */
35
36 /**
37 * \file
38 *
39 * This file contains a module that performs recursive iterative DNS query
40 * processing.
41 */
42
43 #include "config.h"
44 #include "iterator/iterator.h"
45 #include "iterator/iter_utils.h"
46 #include "iterator/iter_hints.h"
47 #include "iterator/iter_fwd.h"
48 #include "iterator/iter_donotq.h"
49 #include "iterator/iter_delegpt.h"
50 #include "iterator/iter_resptype.h"
51 #include "iterator/iter_scrub.h"
52 #include "iterator/iter_priv.h"
53 #include "validator/val_neg.h"
54 #include "services/cache/dns.h"
55 #include "services/cache/infra.h"
56 #include "services/authzone.h"
57 #include "util/module.h"
58 #include "util/netevent.h"
59 #include "util/net_help.h"
60 #include "util/regional.h"
61 #include "util/data/dname.h"
62 #include "util/data/msgencode.h"
63 #include "util/fptr_wlist.h"
64 #include "util/config_file.h"
65 #include "util/random.h"
66 #include "sldns/rrdef.h"
67 #include "sldns/wire2str.h"
68 #include "sldns/str2wire.h"
69 #include "sldns/parseutil.h"
70 #include "sldns/sbuffer.h"
71
72 /* in msec */
73 int UNKNOWN_SERVER_NICENESS = 376;
74 /* in msec */
75 int USEFUL_SERVER_TOP_TIMEOUT = 120000;
76 /* Equals USEFUL_SERVER_TOP_TIMEOUT*4 */
77 int BLACKLIST_PENALTY = (120000*4);
78
79 static void target_count_increase_nx(struct iter_qstate* iq, int num);
80
81 int
iter_init(struct module_env * env,int id)82 iter_init(struct module_env* env, int id)
83 {
84 struct iter_env* iter_env = (struct iter_env*)calloc(1,
85 sizeof(struct iter_env));
86 if(!iter_env) {
87 log_err("malloc failure");
88 return 0;
89 }
90 env->modinfo[id] = (void*)iter_env;
91
92 lock_basic_init(&iter_env->queries_ratelimit_lock);
93 lock_protect(&iter_env->queries_ratelimit_lock,
94 &iter_env->num_queries_ratelimited,
95 sizeof(iter_env->num_queries_ratelimited));
96
97 if(!iter_apply_cfg(iter_env, env->cfg)) {
98 log_err("iterator: could not apply configuration settings.");
99 return 0;
100 }
101
102 return 1;
103 }
104
105 /** delete caps_whitelist element */
106 static void
caps_free(struct rbnode_type * n,void * ATTR_UNUSED (d))107 caps_free(struct rbnode_type* n, void* ATTR_UNUSED(d))
108 {
109 if(n) {
110 free(((struct name_tree_node*)n)->name);
111 free(n);
112 }
113 }
114
115 void
iter_deinit(struct module_env * env,int id)116 iter_deinit(struct module_env* env, int id)
117 {
118 struct iter_env* iter_env;
119 if(!env || !env->modinfo[id])
120 return;
121 iter_env = (struct iter_env*)env->modinfo[id];
122 lock_basic_destroy(&iter_env->queries_ratelimit_lock);
123 free(iter_env->target_fetch_policy);
124 priv_delete(iter_env->priv);
125 donotq_delete(iter_env->donotq);
126 if(iter_env->caps_white) {
127 traverse_postorder(iter_env->caps_white, caps_free, NULL);
128 free(iter_env->caps_white);
129 }
130 free(iter_env);
131 env->modinfo[id] = NULL;
132 }
133
134 /** new query for iterator */
135 static int
iter_new(struct module_qstate * qstate,int id)136 iter_new(struct module_qstate* qstate, int id)
137 {
138 struct iter_qstate* iq = (struct iter_qstate*)regional_alloc(
139 qstate->region, sizeof(struct iter_qstate));
140 qstate->minfo[id] = iq;
141 if(!iq)
142 return 0;
143 memset(iq, 0, sizeof(*iq));
144 iq->state = INIT_REQUEST_STATE;
145 iq->final_state = FINISHED_STATE;
146 iq->an_prepend_list = NULL;
147 iq->an_prepend_last = NULL;
148 iq->ns_prepend_list = NULL;
149 iq->ns_prepend_last = NULL;
150 iq->dp = NULL;
151 iq->depth = 0;
152 iq->num_target_queries = 0;
153 iq->num_current_queries = 0;
154 iq->query_restart_count = 0;
155 iq->referral_count = 0;
156 iq->sent_count = 0;
157 iq->ratelimit_ok = 0;
158 iq->target_count = NULL;
159 iq->dp_target_count = 0;
160 iq->wait_priming_stub = 0;
161 iq->refetch_glue = 0;
162 iq->dnssec_expected = 0;
163 iq->dnssec_lame_query = 0;
164 iq->chase_flags = qstate->query_flags;
165 /* Start with the (current) qname. */
166 iq->qchase = qstate->qinfo;
167 outbound_list_init(&iq->outlist);
168 iq->minimise_count = 0;
169 iq->timeout_count = 0;
170 if (qstate->env->cfg->qname_minimisation)
171 iq->minimisation_state = INIT_MINIMISE_STATE;
172 else
173 iq->minimisation_state = DONOT_MINIMISE_STATE;
174
175 memset(&iq->qinfo_out, 0, sizeof(struct query_info));
176 return 1;
177 }
178
179 /**
180 * Transition to the next state. This can be used to advance a currently
181 * processing event. It cannot be used to reactivate a forEvent.
182 *
183 * @param iq: iterator query state
184 * @param nextstate The state to transition to.
185 * @return true. This is so this can be called as the return value for the
186 * actual process*State() methods. (Transitioning to the next state
187 * implies further processing).
188 */
189 static int
next_state(struct iter_qstate * iq,enum iter_state nextstate)190 next_state(struct iter_qstate* iq, enum iter_state nextstate)
191 {
192 /* If transitioning to a "response" state, make sure that there is a
193 * response */
194 if(iter_state_is_responsestate(nextstate)) {
195 if(iq->response == NULL) {
196 log_err("transitioning to response state sans "
197 "response.");
198 }
199 }
200 iq->state = nextstate;
201 return 1;
202 }
203
204 /**
205 * Transition an event to its final state. Final states always either return
206 * a result up the module chain, or reactivate a dependent event. Which
207 * final state to transition to is set in the module state for the event when
208 * it was created, and depends on the original purpose of the event.
209 *
210 * The response is stored in the qstate->buf buffer.
211 *
212 * @param iq: iterator query state
213 * @return false. This is so this method can be used as the return value for
214 * the processState methods. (Transitioning to the final state
215 */
216 static int
final_state(struct iter_qstate * iq)217 final_state(struct iter_qstate* iq)
218 {
219 return next_state(iq, iq->final_state);
220 }
221
222 /**
223 * Callback routine to handle errors in parent query states
224 * @param qstate: query state that failed.
225 * @param id: module id.
226 * @param super: super state.
227 */
228 static void
error_supers(struct module_qstate * qstate,int id,struct module_qstate * super)229 error_supers(struct module_qstate* qstate, int id, struct module_qstate* super)
230 {
231 struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
232 struct iter_qstate* super_iq = (struct iter_qstate*)super->minfo[id];
233
234 if(qstate->qinfo.qtype == LDNS_RR_TYPE_A ||
235 qstate->qinfo.qtype == LDNS_RR_TYPE_AAAA) {
236 /* mark address as failed. */
237 struct delegpt_ns* dpns = NULL;
238 super_iq->num_target_queries--;
239 if(super_iq->dp)
240 dpns = delegpt_find_ns(super_iq->dp,
241 qstate->qinfo.qname, qstate->qinfo.qname_len);
242 if(!dpns) {
243 /* not interested */
244 /* this can happen, for eg. qname minimisation asked
245 * for an NXDOMAIN to be validated, and used qtype
246 * A for that, and the error of that, the name, is
247 * not listed in super_iq->dp */
248 verbose(VERB_ALGO, "subq error, but not interested");
249 log_query_info(VERB_ALGO, "superq", &super->qinfo);
250 return;
251 } else {
252 /* see if the failure did get (parent-lame) info */
253 if(!cache_fill_missing(super->env, super_iq->qchase.qclass,
254 super->region, super_iq->dp))
255 log_err("out of memory adding missing");
256 }
257 delegpt_mark_neg(dpns, qstate->qinfo.qtype);
258 dpns->resolved = 1; /* mark as failed */
259 if((dpns->got4 == 2 || !ie->supports_ipv4) &&
260 (dpns->got6 == 2 || !ie->supports_ipv6)) {
261 target_count_increase_nx(super_iq, 1);
262 }
263 }
264 if(qstate->qinfo.qtype == LDNS_RR_TYPE_NS) {
265 /* prime failed to get delegation */
266 super_iq->dp = NULL;
267 }
268 /* evaluate targets again */
269 super_iq->state = QUERYTARGETS_STATE;
270 /* super becomes runnable, and will process this change */
271 }
272
273 /**
274 * Return an error to the client
275 * @param qstate: our query state
276 * @param id: module id
277 * @param rcode: error code (DNS errcode).
278 * @return: 0 for use by caller, to make notation easy, like:
279 * return error_response(..).
280 */
281 static int
error_response(struct module_qstate * qstate,int id,int rcode)282 error_response(struct module_qstate* qstate, int id, int rcode)
283 {
284 verbose(VERB_QUERY, "return error response %s",
285 sldns_lookup_by_id(sldns_rcodes, rcode)?
286 sldns_lookup_by_id(sldns_rcodes, rcode)->name:"??");
287 qstate->return_rcode = rcode;
288 qstate->return_msg = NULL;
289 qstate->ext_state[id] = module_finished;
290 return 0;
291 }
292
293 /**
294 * Return an error to the client and cache the error code in the
295 * message cache (so per qname, qtype, qclass).
296 * @param qstate: our query state
297 * @param id: module id
298 * @param rcode: error code (DNS errcode).
299 * @return: 0 for use by caller, to make notation easy, like:
300 * return error_response(..).
301 */
302 static int
error_response_cache(struct module_qstate * qstate,int id,int rcode)303 error_response_cache(struct module_qstate* qstate, int id, int rcode)
304 {
305 if(!qstate->no_cache_store) {
306 /* store in cache */
307 struct reply_info err;
308 if(qstate->prefetch_leeway > NORR_TTL) {
309 verbose(VERB_ALGO, "error response for prefetch in cache");
310 /* attempt to adjust the cache entry prefetch */
311 if(dns_cache_prefetch_adjust(qstate->env, &qstate->qinfo,
312 NORR_TTL, qstate->query_flags))
313 return error_response(qstate, id, rcode);
314 /* if that fails (not in cache), fall through to store err */
315 }
316 if(qstate->env->cfg->serve_expired) {
317 /* if serving expired contents, and such content is
318 * already available, don't overwrite this servfail */
319 struct msgreply_entry* msg;
320 if((msg=msg_cache_lookup(qstate->env,
321 qstate->qinfo.qname, qstate->qinfo.qname_len,
322 qstate->qinfo.qtype, qstate->qinfo.qclass,
323 qstate->query_flags, 0,
324 qstate->env->cfg->serve_expired_ttl_reset))
325 != NULL) {
326 if(qstate->env->cfg->serve_expired_ttl_reset) {
327 struct reply_info* rep =
328 (struct reply_info*)msg->entry.data;
329 if(rep && *qstate->env->now +
330 qstate->env->cfg->serve_expired_ttl >
331 rep->serve_expired_ttl) {
332 rep->serve_expired_ttl =
333 *qstate->env->now +
334 qstate->env->cfg->serve_expired_ttl;
335 }
336 }
337 lock_rw_unlock(&msg->entry.lock);
338 return error_response(qstate, id, rcode);
339 }
340 /* serving expired contents, but nothing is cached
341 * at all, so the servfail cache entry is useful
342 * (stops waste of time on this servfail NORR_TTL) */
343 } else {
344 /* don't overwrite existing (non-expired) data in
345 * cache with a servfail */
346 struct msgreply_entry* msg;
347 if((msg=msg_cache_lookup(qstate->env,
348 qstate->qinfo.qname, qstate->qinfo.qname_len,
349 qstate->qinfo.qtype, qstate->qinfo.qclass,
350 qstate->query_flags, *qstate->env->now, 0))
351 != NULL) {
352 struct reply_info* rep = (struct reply_info*)
353 msg->entry.data;
354 if(FLAGS_GET_RCODE(rep->flags) ==
355 LDNS_RCODE_NOERROR ||
356 FLAGS_GET_RCODE(rep->flags) ==
357 LDNS_RCODE_NXDOMAIN) {
358 /* we have a good entry,
359 * don't overwrite */
360 lock_rw_unlock(&msg->entry.lock);
361 return error_response(qstate, id, rcode);
362 }
363 lock_rw_unlock(&msg->entry.lock);
364 }
365
366 }
367 memset(&err, 0, sizeof(err));
368 err.flags = (uint16_t)(BIT_QR | BIT_RA);
369 FLAGS_SET_RCODE(err.flags, rcode);
370 err.qdcount = 1;
371 err.ttl = NORR_TTL;
372 err.prefetch_ttl = PREFETCH_TTL_CALC(err.ttl);
373 err.serve_expired_ttl = NORR_TTL;
374 /* do not waste time trying to validate this servfail */
375 err.security = sec_status_indeterminate;
376 verbose(VERB_ALGO, "store error response in message cache");
377 iter_dns_store(qstate->env, &qstate->qinfo, &err, 0, 0, 0, NULL,
378 qstate->query_flags, qstate->qstarttime);
379 }
380 return error_response(qstate, id, rcode);
381 }
382
383 /** check if prepend item is duplicate item */
384 static int
prepend_is_duplicate(struct ub_packed_rrset_key ** sets,size_t to,struct ub_packed_rrset_key * dup)385 prepend_is_duplicate(struct ub_packed_rrset_key** sets, size_t to,
386 struct ub_packed_rrset_key* dup)
387 {
388 size_t i;
389 for(i=0; i<to; i++) {
390 if(sets[i]->rk.type == dup->rk.type &&
391 sets[i]->rk.rrset_class == dup->rk.rrset_class &&
392 sets[i]->rk.dname_len == dup->rk.dname_len &&
393 query_dname_compare(sets[i]->rk.dname, dup->rk.dname)
394 == 0)
395 return 1;
396 }
397 return 0;
398 }
399
400 /** prepend the prepend list in the answer and authority section of dns_msg */
401 static int
iter_prepend(struct iter_qstate * iq,struct dns_msg * msg,struct regional * region)402 iter_prepend(struct iter_qstate* iq, struct dns_msg* msg,
403 struct regional* region)
404 {
405 struct iter_prep_list* p;
406 struct ub_packed_rrset_key** sets;
407 size_t num_an = 0, num_ns = 0;;
408 for(p = iq->an_prepend_list; p; p = p->next)
409 num_an++;
410 for(p = iq->ns_prepend_list; p; p = p->next)
411 num_ns++;
412 if(num_an + num_ns == 0)
413 return 1;
414 verbose(VERB_ALGO, "prepending %d rrsets", (int)num_an + (int)num_ns);
415 if(num_an > RR_COUNT_MAX || num_ns > RR_COUNT_MAX ||
416 msg->rep->rrset_count > RR_COUNT_MAX) return 0; /* overflow */
417 sets = regional_alloc(region, (num_an+num_ns+msg->rep->rrset_count) *
418 sizeof(struct ub_packed_rrset_key*));
419 if(!sets)
420 return 0;
421 /* ANSWER section */
422 num_an = 0;
423 for(p = iq->an_prepend_list; p; p = p->next) {
424 sets[num_an++] = p->rrset;
425 if(ub_packed_rrset_ttl(p->rrset) < msg->rep->ttl)
426 msg->rep->ttl = ub_packed_rrset_ttl(p->rrset);
427 }
428 memcpy(sets+num_an, msg->rep->rrsets, msg->rep->an_numrrsets *
429 sizeof(struct ub_packed_rrset_key*));
430 /* AUTH section */
431 num_ns = 0;
432 for(p = iq->ns_prepend_list; p; p = p->next) {
433 if(prepend_is_duplicate(sets+msg->rep->an_numrrsets+num_an,
434 num_ns, p->rrset) || prepend_is_duplicate(
435 msg->rep->rrsets+msg->rep->an_numrrsets,
436 msg->rep->ns_numrrsets, p->rrset))
437 continue;
438 sets[msg->rep->an_numrrsets + num_an + num_ns++] = p->rrset;
439 if(ub_packed_rrset_ttl(p->rrset) < msg->rep->ttl)
440 msg->rep->ttl = ub_packed_rrset_ttl(p->rrset);
441 }
442 memcpy(sets + num_an + msg->rep->an_numrrsets + num_ns,
443 msg->rep->rrsets + msg->rep->an_numrrsets,
444 (msg->rep->ns_numrrsets + msg->rep->ar_numrrsets) *
445 sizeof(struct ub_packed_rrset_key*));
446
447 /* NXDOMAIN rcode can stay if we prepended DNAME/CNAMEs, because
448 * this is what recursors should give. */
449 msg->rep->rrset_count += num_an + num_ns;
450 msg->rep->an_numrrsets += num_an;
451 msg->rep->ns_numrrsets += num_ns;
452 msg->rep->rrsets = sets;
453 return 1;
454 }
455
456 /**
457 * Find rrset in ANSWER prepend list.
458 * to avoid duplicate DNAMEs when a DNAME is traversed twice.
459 * @param iq: iterator query state.
460 * @param rrset: rrset to add.
461 * @return false if not found
462 */
463 static int
iter_find_rrset_in_prepend_answer(struct iter_qstate * iq,struct ub_packed_rrset_key * rrset)464 iter_find_rrset_in_prepend_answer(struct iter_qstate* iq,
465 struct ub_packed_rrset_key* rrset)
466 {
467 struct iter_prep_list* p = iq->an_prepend_list;
468 while(p) {
469 if(ub_rrset_compare(p->rrset, rrset) == 0 &&
470 rrsetdata_equal((struct packed_rrset_data*)p->rrset
471 ->entry.data, (struct packed_rrset_data*)rrset
472 ->entry.data))
473 return 1;
474 p = p->next;
475 }
476 return 0;
477 }
478
479 /**
480 * Add rrset to ANSWER prepend list
481 * @param qstate: query state.
482 * @param iq: iterator query state.
483 * @param rrset: rrset to add.
484 * @return false on failure (malloc).
485 */
486 static int
iter_add_prepend_answer(struct module_qstate * qstate,struct iter_qstate * iq,struct ub_packed_rrset_key * rrset)487 iter_add_prepend_answer(struct module_qstate* qstate, struct iter_qstate* iq,
488 struct ub_packed_rrset_key* rrset)
489 {
490 struct iter_prep_list* p = (struct iter_prep_list*)regional_alloc(
491 qstate->region, sizeof(struct iter_prep_list));
492 if(!p)
493 return 0;
494 p->rrset = rrset;
495 p->next = NULL;
496 /* add at end */
497 if(iq->an_prepend_last)
498 iq->an_prepend_last->next = p;
499 else iq->an_prepend_list = p;
500 iq->an_prepend_last = p;
501 return 1;
502 }
503
504 /**
505 * Add rrset to AUTHORITY prepend list
506 * @param qstate: query state.
507 * @param iq: iterator query state.
508 * @param rrset: rrset to add.
509 * @return false on failure (malloc).
510 */
511 static int
iter_add_prepend_auth(struct module_qstate * qstate,struct iter_qstate * iq,struct ub_packed_rrset_key * rrset)512 iter_add_prepend_auth(struct module_qstate* qstate, struct iter_qstate* iq,
513 struct ub_packed_rrset_key* rrset)
514 {
515 struct iter_prep_list* p = (struct iter_prep_list*)regional_alloc(
516 qstate->region, sizeof(struct iter_prep_list));
517 if(!p)
518 return 0;
519 p->rrset = rrset;
520 p->next = NULL;
521 /* add at end */
522 if(iq->ns_prepend_last)
523 iq->ns_prepend_last->next = p;
524 else iq->ns_prepend_list = p;
525 iq->ns_prepend_last = p;
526 return 1;
527 }
528
529 /**
530 * Given a CNAME response (defined as a response containing a CNAME or DNAME
531 * that does not answer the request), process the response, modifying the
532 * state as necessary. This follows the CNAME/DNAME chain and returns the
533 * final query name.
534 *
535 * sets the new query name, after following the CNAME/DNAME chain.
536 * @param qstate: query state.
537 * @param iq: iterator query state.
538 * @param msg: the response.
539 * @param mname: returned target new query name.
540 * @param mname_len: length of mname.
541 * @return false on (malloc) error.
542 */
543 static int
handle_cname_response(struct module_qstate * qstate,struct iter_qstate * iq,struct dns_msg * msg,uint8_t ** mname,size_t * mname_len)544 handle_cname_response(struct module_qstate* qstate, struct iter_qstate* iq,
545 struct dns_msg* msg, uint8_t** mname, size_t* mname_len)
546 {
547 size_t i;
548 /* Start with the (current) qname. */
549 *mname = iq->qchase.qname;
550 *mname_len = iq->qchase.qname_len;
551
552 /* Iterate over the ANSWER rrsets in order, looking for CNAMEs and
553 * DNAMES. */
554 for(i=0; i<msg->rep->an_numrrsets; i++) {
555 struct ub_packed_rrset_key* r = msg->rep->rrsets[i];
556 /* If there is a (relevant) DNAME, add it to the list.
557 * We always expect there to be CNAME that was generated
558 * by this DNAME following, so we don't process the DNAME
559 * directly. */
560 if(ntohs(r->rk.type) == LDNS_RR_TYPE_DNAME &&
561 dname_strict_subdomain_c(*mname, r->rk.dname) &&
562 !iter_find_rrset_in_prepend_answer(iq, r)) {
563 if(!iter_add_prepend_answer(qstate, iq, r))
564 return 0;
565 continue;
566 }
567
568 if(ntohs(r->rk.type) == LDNS_RR_TYPE_CNAME &&
569 query_dname_compare(*mname, r->rk.dname) == 0 &&
570 !iter_find_rrset_in_prepend_answer(iq, r)) {
571 /* Add this relevant CNAME rrset to the prepend list.*/
572 if(!iter_add_prepend_answer(qstate, iq, r))
573 return 0;
574 get_cname_target(r, mname, mname_len);
575 }
576
577 /* Other rrsets in the section are ignored. */
578 }
579 /* add authority rrsets to authority prepend, for wildcarded CNAMEs */
580 for(i=msg->rep->an_numrrsets; i<msg->rep->an_numrrsets +
581 msg->rep->ns_numrrsets; i++) {
582 struct ub_packed_rrset_key* r = msg->rep->rrsets[i];
583 /* only add NSEC/NSEC3, as they may be needed for validation */
584 if(ntohs(r->rk.type) == LDNS_RR_TYPE_NSEC ||
585 ntohs(r->rk.type) == LDNS_RR_TYPE_NSEC3) {
586 if(!iter_add_prepend_auth(qstate, iq, r))
587 return 0;
588 }
589 }
590 return 1;
591 }
592
593 /** add response specific error information for log servfail */
594 static void
errinf_reply(struct module_qstate * qstate,struct iter_qstate * iq)595 errinf_reply(struct module_qstate* qstate, struct iter_qstate* iq)
596 {
597 if(qstate->env->cfg->val_log_level < 2 && !qstate->env->cfg->log_servfail)
598 return;
599 if((qstate->reply && qstate->reply->addrlen != 0) ||
600 (iq->fail_reply && iq->fail_reply->addrlen != 0)) {
601 char from[256], frm[512];
602 if(qstate->reply && qstate->reply->addrlen != 0)
603 addr_to_str(&qstate->reply->addr, qstate->reply->addrlen,
604 from, sizeof(from));
605 else
606 addr_to_str(&iq->fail_reply->addr, iq->fail_reply->addrlen,
607 from, sizeof(from));
608 snprintf(frm, sizeof(frm), "from %s", from);
609 errinf(qstate, frm);
610 }
611 if(iq->scrub_failures || iq->parse_failures) {
612 if(iq->scrub_failures)
613 errinf(qstate, "upstream response failed scrub");
614 if(iq->parse_failures)
615 errinf(qstate, "could not parse upstream response");
616 } else if(iq->response == NULL && iq->timeout_count != 0) {
617 errinf(qstate, "upstream server timeout");
618 } else if(iq->response == NULL) {
619 errinf(qstate, "no server to query");
620 if(iq->dp) {
621 if(iq->dp->target_list == NULL)
622 errinf(qstate, "no addresses for nameservers");
623 else errinf(qstate, "nameserver addresses not usable");
624 if(iq->dp->nslist == NULL)
625 errinf(qstate, "have no nameserver names");
626 if(iq->dp->bogus)
627 errinf(qstate, "NS record was dnssec bogus");
628 }
629 }
630 if(iq->response && iq->response->rep) {
631 if(FLAGS_GET_RCODE(iq->response->rep->flags) != 0) {
632 char rcode[256], rc[32];
633 (void)sldns_wire2str_rcode_buf(
634 FLAGS_GET_RCODE(iq->response->rep->flags),
635 rc, sizeof(rc));
636 snprintf(rcode, sizeof(rcode), "got %s", rc);
637 errinf(qstate, rcode);
638 } else {
639 /* rcode NOERROR */
640 if(iq->response->rep->an_numrrsets == 0) {
641 errinf(qstate, "nodata answer");
642 }
643 }
644 }
645 }
646
647 /** see if last resort is possible - does config allow queries to parent */
648 static int
can_have_last_resort(struct module_env * env,uint8_t * nm,size_t nmlen,uint16_t qclass,struct delegpt ** retdp)649 can_have_last_resort(struct module_env* env, uint8_t* nm, size_t nmlen,
650 uint16_t qclass, struct delegpt** retdp)
651 {
652 struct delegpt* fwddp;
653 struct iter_hints_stub* stub;
654 int labs = dname_count_labels(nm);
655 /* do not process a last resort (the parent side) if a stub
656 * or forward is configured, because we do not want to go 'above'
657 * the configured servers */
658 if(!dname_is_root(nm) && (stub = (struct iter_hints_stub*)
659 name_tree_find(&env->hints->tree, nm, nmlen, labs, qclass)) &&
660 /* has_parent side is turned off for stub_first, where we
661 * are allowed to go to the parent */
662 stub->dp->has_parent_side_NS) {
663 if(retdp) *retdp = stub->dp;
664 return 0;
665 }
666 if((fwddp = forwards_find(env->fwds, nm, qclass)) &&
667 /* has_parent_side is turned off for forward_first, where
668 * we are allowed to go to the parent */
669 fwddp->has_parent_side_NS) {
670 if(retdp) *retdp = fwddp;
671 return 0;
672 }
673 return 1;
674 }
675
676 /** see if target name is caps-for-id whitelisted */
677 static int
is_caps_whitelisted(struct iter_env * ie,struct iter_qstate * iq)678 is_caps_whitelisted(struct iter_env* ie, struct iter_qstate* iq)
679 {
680 if(!ie->caps_white) return 0; /* no whitelist, or no capsforid */
681 return name_tree_lookup(ie->caps_white, iq->qchase.qname,
682 iq->qchase.qname_len, dname_count_labels(iq->qchase.qname),
683 iq->qchase.qclass) != NULL;
684 }
685
686 /**
687 * Create target count structure for this query. This is always explicitly
688 * created for the parent query.
689 */
690 static void
target_count_create(struct iter_qstate * iq)691 target_count_create(struct iter_qstate* iq)
692 {
693 if(!iq->target_count) {
694 iq->target_count = (int*)calloc(TARGET_COUNT_MAX, sizeof(int));
695 /* if calloc fails we simply do not track this number */
696 if(iq->target_count) {
697 iq->target_count[TARGET_COUNT_REF] = 1;
698 iq->nxns_dp = (uint8_t**)calloc(1, sizeof(uint8_t*));
699 }
700 }
701 }
702
703 static void
target_count_increase(struct iter_qstate * iq,int num)704 target_count_increase(struct iter_qstate* iq, int num)
705 {
706 target_count_create(iq);
707 if(iq->target_count)
708 iq->target_count[TARGET_COUNT_QUERIES] += num;
709 iq->dp_target_count++;
710 }
711
712 static void
target_count_increase_nx(struct iter_qstate * iq,int num)713 target_count_increase_nx(struct iter_qstate* iq, int num)
714 {
715 target_count_create(iq);
716 if(iq->target_count)
717 iq->target_count[TARGET_COUNT_NX] += num;
718 }
719
720 /**
721 * Generate a subrequest.
722 * Generate a local request event. Local events are tied to this module, and
723 * have a corresponding (first tier) event that is waiting for this event to
724 * resolve to continue.
725 *
726 * @param qname The query name for this request.
727 * @param qnamelen length of qname
728 * @param qtype The query type for this request.
729 * @param qclass The query class for this request.
730 * @param qstate The event that is generating this event.
731 * @param id: module id.
732 * @param iq: The iterator state that is generating this event.
733 * @param initial_state The initial response state (normally this
734 * is QUERY_RESP_STATE, unless it is known that the request won't
735 * need iterative processing
736 * @param finalstate The final state for the response to this request.
737 * @param subq_ret: if newly allocated, the subquerystate, or NULL if it does
738 * not need initialisation.
739 * @param v: if true, validation is done on the subquery.
740 * @param detached: true if this qstate should not attach to the subquery
741 * @return false on error (malloc).
742 */
743 static int
generate_sub_request(uint8_t * qname,size_t qnamelen,uint16_t qtype,uint16_t qclass,struct module_qstate * qstate,int id,struct iter_qstate * iq,enum iter_state initial_state,enum iter_state finalstate,struct module_qstate ** subq_ret,int v,int detached)744 generate_sub_request(uint8_t* qname, size_t qnamelen, uint16_t qtype,
745 uint16_t qclass, struct module_qstate* qstate, int id,
746 struct iter_qstate* iq, enum iter_state initial_state,
747 enum iter_state finalstate, struct module_qstate** subq_ret, int v,
748 int detached)
749 {
750 struct module_qstate* subq = NULL;
751 struct iter_qstate* subiq = NULL;
752 uint16_t qflags = 0; /* OPCODE QUERY, no flags */
753 struct query_info qinf;
754 int prime = (finalstate == PRIME_RESP_STATE)?1:0;
755 int valrec = 0;
756 qinf.qname = qname;
757 qinf.qname_len = qnamelen;
758 qinf.qtype = qtype;
759 qinf.qclass = qclass;
760 qinf.local_alias = NULL;
761
762 /* RD should be set only when sending the query back through the INIT
763 * state. */
764 if(initial_state == INIT_REQUEST_STATE)
765 qflags |= BIT_RD;
766 /* We set the CD flag so we can send this through the "head" of
767 * the resolution chain, which might have a validator. We are
768 * uninterested in validating things not on the direct resolution
769 * path. */
770 if(!v) {
771 qflags |= BIT_CD;
772 valrec = 1;
773 }
774
775 if(detached) {
776 struct mesh_state* sub = NULL;
777 fptr_ok(fptr_whitelist_modenv_add_sub(
778 qstate->env->add_sub));
779 if(!(*qstate->env->add_sub)(qstate, &qinf,
780 qflags, prime, valrec, &subq, &sub)){
781 return 0;
782 }
783 }
784 else {
785 /* attach subquery, lookup existing or make a new one */
786 fptr_ok(fptr_whitelist_modenv_attach_sub(
787 qstate->env->attach_sub));
788 if(!(*qstate->env->attach_sub)(qstate, &qinf, qflags, prime,
789 valrec, &subq)) {
790 return 0;
791 }
792 }
793 *subq_ret = subq;
794 if(subq) {
795 /* initialise the new subquery */
796 subq->curmod = id;
797 subq->ext_state[id] = module_state_initial;
798 subq->minfo[id] = regional_alloc(subq->region,
799 sizeof(struct iter_qstate));
800 if(!subq->minfo[id]) {
801 log_err("init subq: out of memory");
802 fptr_ok(fptr_whitelist_modenv_kill_sub(
803 qstate->env->kill_sub));
804 (*qstate->env->kill_sub)(subq);
805 return 0;
806 }
807 subiq = (struct iter_qstate*)subq->minfo[id];
808 memset(subiq, 0, sizeof(*subiq));
809 subiq->num_target_queries = 0;
810 target_count_create(iq);
811 subiq->target_count = iq->target_count;
812 if(iq->target_count) {
813 iq->target_count[TARGET_COUNT_REF] ++; /* extra reference */
814 subiq->nxns_dp = iq->nxns_dp;
815 }
816 subiq->dp_target_count = 0;
817 subiq->num_current_queries = 0;
818 subiq->depth = iq->depth+1;
819 outbound_list_init(&subiq->outlist);
820 subiq->state = initial_state;
821 subiq->final_state = finalstate;
822 subiq->qchase = subq->qinfo;
823 subiq->chase_flags = subq->query_flags;
824 subiq->refetch_glue = 0;
825 if(qstate->env->cfg->qname_minimisation)
826 subiq->minimisation_state = INIT_MINIMISE_STATE;
827 else
828 subiq->minimisation_state = DONOT_MINIMISE_STATE;
829 memset(&subiq->qinfo_out, 0, sizeof(struct query_info));
830 }
831 return 1;
832 }
833
834 /**
835 * Generate and send a root priming request.
836 * @param qstate: the qtstate that triggered the need to prime.
837 * @param iq: iterator query state.
838 * @param id: module id.
839 * @param qclass: the class to prime.
840 * @return 0 on failure
841 */
842 static int
prime_root(struct module_qstate * qstate,struct iter_qstate * iq,int id,uint16_t qclass)843 prime_root(struct module_qstate* qstate, struct iter_qstate* iq, int id,
844 uint16_t qclass)
845 {
846 struct delegpt* dp;
847 struct module_qstate* subq;
848 verbose(VERB_DETAIL, "priming . %s NS",
849 sldns_lookup_by_id(sldns_rr_classes, (int)qclass)?
850 sldns_lookup_by_id(sldns_rr_classes, (int)qclass)->name:"??");
851 dp = hints_lookup_root(qstate->env->hints, qclass);
852 if(!dp) {
853 verbose(VERB_ALGO, "Cannot prime due to lack of hints");
854 return 0;
855 }
856 /* Priming requests start at the QUERYTARGETS state, skipping
857 * the normal INIT state logic (which would cause an infloop). */
858 if(!generate_sub_request((uint8_t*)"\000", 1, LDNS_RR_TYPE_NS,
859 qclass, qstate, id, iq, QUERYTARGETS_STATE, PRIME_RESP_STATE,
860 &subq, 0, 0)) {
861 verbose(VERB_ALGO, "could not prime root");
862 return 0;
863 }
864 if(subq) {
865 struct iter_qstate* subiq =
866 (struct iter_qstate*)subq->minfo[id];
867 /* Set the initial delegation point to the hint.
868 * copy dp, it is now part of the root prime query.
869 * dp was part of in the fixed hints structure. */
870 subiq->dp = delegpt_copy(dp, subq->region);
871 if(!subiq->dp) {
872 log_err("out of memory priming root, copydp");
873 fptr_ok(fptr_whitelist_modenv_kill_sub(
874 qstate->env->kill_sub));
875 (*qstate->env->kill_sub)(subq);
876 return 0;
877 }
878 /* there should not be any target queries. */
879 subiq->num_target_queries = 0;
880 subiq->dnssec_expected = iter_indicates_dnssec(
881 qstate->env, subiq->dp, NULL, subq->qinfo.qclass);
882 }
883
884 /* this module stops, our submodule starts, and does the query. */
885 qstate->ext_state[id] = module_wait_subquery;
886 return 1;
887 }
888
889 /**
890 * Generate and process a stub priming request. This method tests for the
891 * need to prime a stub zone, so it is safe to call for every request.
892 *
893 * @param qstate: the qtstate that triggered the need to prime.
894 * @param iq: iterator query state.
895 * @param id: module id.
896 * @param qname: request name.
897 * @param qclass: request class.
898 * @return true if a priming subrequest was made, false if not. The will only
899 * issue a priming request if it detects an unprimed stub.
900 * Uses value of 2 to signal during stub-prime in root-prime situation
901 * that a noprime-stub is available and resolution can continue.
902 */
903 static int
prime_stub(struct module_qstate * qstate,struct iter_qstate * iq,int id,uint8_t * qname,uint16_t qclass)904 prime_stub(struct module_qstate* qstate, struct iter_qstate* iq, int id,
905 uint8_t* qname, uint16_t qclass)
906 {
907 /* Lookup the stub hint. This will return null if the stub doesn't
908 * need to be re-primed. */
909 struct iter_hints_stub* stub;
910 struct delegpt* stub_dp;
911 struct module_qstate* subq;
912
913 if(!qname) return 0;
914 stub = hints_lookup_stub(qstate->env->hints, qname, qclass, iq->dp);
915 /* The stub (if there is one) does not need priming. */
916 if(!stub)
917 return 0;
918 stub_dp = stub->dp;
919 /* if we have an auth_zone dp, and stub is equal, don't prime stub
920 * yet, unless we want to fallback and avoid the auth_zone */
921 if(!iq->auth_zone_avoid && iq->dp && iq->dp->auth_dp &&
922 query_dname_compare(iq->dp->name, stub_dp->name) == 0)
923 return 0;
924
925 /* is it a noprime stub (always use) */
926 if(stub->noprime) {
927 int r = 0;
928 if(iq->dp == NULL) r = 2;
929 /* copy the dp out of the fixed hints structure, so that
930 * it can be changed when servicing this query */
931 iq->dp = delegpt_copy(stub_dp, qstate->region);
932 if(!iq->dp) {
933 log_err("out of memory priming stub");
934 errinf(qstate, "malloc failure, priming stub");
935 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
936 return 1; /* return 1 to make module stop, with error */
937 }
938 log_nametypeclass(VERB_DETAIL, "use stub", stub_dp->name,
939 LDNS_RR_TYPE_NS, qclass);
940 return r;
941 }
942
943 /* Otherwise, we need to (re)prime the stub. */
944 log_nametypeclass(VERB_DETAIL, "priming stub", stub_dp->name,
945 LDNS_RR_TYPE_NS, qclass);
946
947 /* Stub priming events start at the QUERYTARGETS state to avoid the
948 * redundant INIT state processing. */
949 if(!generate_sub_request(stub_dp->name, stub_dp->namelen,
950 LDNS_RR_TYPE_NS, qclass, qstate, id, iq,
951 QUERYTARGETS_STATE, PRIME_RESP_STATE, &subq, 0, 0)) {
952 verbose(VERB_ALGO, "could not prime stub");
953 errinf(qstate, "could not generate lookup for stub prime");
954 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
955 return 1; /* return 1 to make module stop, with error */
956 }
957 if(subq) {
958 struct iter_qstate* subiq =
959 (struct iter_qstate*)subq->minfo[id];
960
961 /* Set the initial delegation point to the hint. */
962 /* make copy to avoid use of stub dp by different qs/threads */
963 subiq->dp = delegpt_copy(stub_dp, subq->region);
964 if(!subiq->dp) {
965 log_err("out of memory priming stub, copydp");
966 fptr_ok(fptr_whitelist_modenv_kill_sub(
967 qstate->env->kill_sub));
968 (*qstate->env->kill_sub)(subq);
969 errinf(qstate, "malloc failure, in stub prime");
970 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
971 return 1; /* return 1 to make module stop, with error */
972 }
973 /* there should not be any target queries -- although there
974 * wouldn't be anyway, since stub hints never have
975 * missing targets. */
976 subiq->num_target_queries = 0;
977 subiq->wait_priming_stub = 1;
978 subiq->dnssec_expected = iter_indicates_dnssec(
979 qstate->env, subiq->dp, NULL, subq->qinfo.qclass);
980 }
981
982 /* this module stops, our submodule starts, and does the query. */
983 qstate->ext_state[id] = module_wait_subquery;
984 return 1;
985 }
986
987 /**
988 * Generate a delegation point for an auth zone (unless cached dp is better)
989 * false on alloc failure.
990 */
991 static int
auth_zone_delegpt(struct module_qstate * qstate,struct iter_qstate * iq,uint8_t * delname,size_t delnamelen)992 auth_zone_delegpt(struct module_qstate* qstate, struct iter_qstate* iq,
993 uint8_t* delname, size_t delnamelen)
994 {
995 struct auth_zone* z;
996 if(iq->auth_zone_avoid)
997 return 1;
998 if(!delname) {
999 delname = iq->qchase.qname;
1000 delnamelen = iq->qchase.qname_len;
1001 }
1002 lock_rw_rdlock(&qstate->env->auth_zones->lock);
1003 z = auth_zones_find_zone(qstate->env->auth_zones, delname, delnamelen,
1004 qstate->qinfo.qclass);
1005 if(!z) {
1006 lock_rw_unlock(&qstate->env->auth_zones->lock);
1007 return 1;
1008 }
1009 lock_rw_rdlock(&z->lock);
1010 lock_rw_unlock(&qstate->env->auth_zones->lock);
1011 if(z->for_upstream) {
1012 if(iq->dp && query_dname_compare(z->name, iq->dp->name) == 0
1013 && iq->dp->auth_dp && qstate->blacklist &&
1014 z->fallback_enabled) {
1015 /* cache is blacklisted and fallback, and we
1016 * already have an auth_zone dp */
1017 if(verbosity>=VERB_ALGO) {
1018 char buf[255+1];
1019 dname_str(z->name, buf);
1020 verbose(VERB_ALGO, "auth_zone %s "
1021 "fallback because cache blacklisted",
1022 buf);
1023 }
1024 lock_rw_unlock(&z->lock);
1025 iq->dp = NULL;
1026 return 1;
1027 }
1028 if(iq->dp==NULL || dname_subdomain_c(z->name, iq->dp->name)) {
1029 struct delegpt* dp;
1030 if(qstate->blacklist && z->fallback_enabled) {
1031 /* cache is blacklisted because of a DNSSEC
1032 * validation failure, and the zone allows
1033 * fallback to the internet, query there. */
1034 if(verbosity>=VERB_ALGO) {
1035 char buf[255+1];
1036 dname_str(z->name, buf);
1037 verbose(VERB_ALGO, "auth_zone %s "
1038 "fallback because cache blacklisted",
1039 buf);
1040 }
1041 lock_rw_unlock(&z->lock);
1042 return 1;
1043 }
1044 dp = (struct delegpt*)regional_alloc_zero(
1045 qstate->region, sizeof(*dp));
1046 if(!dp) {
1047 log_err("alloc failure");
1048 if(z->fallback_enabled) {
1049 lock_rw_unlock(&z->lock);
1050 return 1; /* just fallback */
1051 }
1052 lock_rw_unlock(&z->lock);
1053 errinf(qstate, "malloc failure");
1054 return 0;
1055 }
1056 dp->name = regional_alloc_init(qstate->region,
1057 z->name, z->namelen);
1058 if(!dp->name) {
1059 log_err("alloc failure");
1060 if(z->fallback_enabled) {
1061 lock_rw_unlock(&z->lock);
1062 return 1; /* just fallback */
1063 }
1064 lock_rw_unlock(&z->lock);
1065 errinf(qstate, "malloc failure");
1066 return 0;
1067 }
1068 dp->namelen = z->namelen;
1069 dp->namelabs = z->namelabs;
1070 dp->auth_dp = 1;
1071 iq->dp = dp;
1072 }
1073 }
1074
1075 lock_rw_unlock(&z->lock);
1076 return 1;
1077 }
1078
1079 /**
1080 * Generate A and AAAA checks for glue that is in-zone for the referral
1081 * we just got to obtain authoritative information on the addresses.
1082 *
1083 * @param qstate: the qtstate that triggered the need to prime.
1084 * @param iq: iterator query state.
1085 * @param id: module id.
1086 */
1087 static void
generate_a_aaaa_check(struct module_qstate * qstate,struct iter_qstate * iq,int id)1088 generate_a_aaaa_check(struct module_qstate* qstate, struct iter_qstate* iq,
1089 int id)
1090 {
1091 struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
1092 struct module_qstate* subq;
1093 size_t i;
1094 struct reply_info* rep = iq->response->rep;
1095 struct ub_packed_rrset_key* s;
1096 log_assert(iq->dp);
1097
1098 if(iq->depth == ie->max_dependency_depth)
1099 return;
1100 /* walk through additional, and check if in-zone,
1101 * only relevant A, AAAA are left after scrub anyway */
1102 for(i=rep->an_numrrsets+rep->ns_numrrsets; i<rep->rrset_count; i++) {
1103 s = rep->rrsets[i];
1104 /* check *ALL* addresses that are transmitted in additional*/
1105 /* is it an address ? */
1106 if( !(ntohs(s->rk.type)==LDNS_RR_TYPE_A ||
1107 ntohs(s->rk.type)==LDNS_RR_TYPE_AAAA)) {
1108 continue;
1109 }
1110 /* is this query the same as the A/AAAA check for it */
1111 if(qstate->qinfo.qtype == ntohs(s->rk.type) &&
1112 qstate->qinfo.qclass == ntohs(s->rk.rrset_class) &&
1113 query_dname_compare(qstate->qinfo.qname,
1114 s->rk.dname)==0 &&
1115 (qstate->query_flags&BIT_RD) &&
1116 !(qstate->query_flags&BIT_CD))
1117 continue;
1118
1119 /* generate subrequest for it */
1120 log_nametypeclass(VERB_ALGO, "schedule addr fetch",
1121 s->rk.dname, ntohs(s->rk.type),
1122 ntohs(s->rk.rrset_class));
1123 if(!generate_sub_request(s->rk.dname, s->rk.dname_len,
1124 ntohs(s->rk.type), ntohs(s->rk.rrset_class),
1125 qstate, id, iq,
1126 INIT_REQUEST_STATE, FINISHED_STATE, &subq, 1, 0)) {
1127 verbose(VERB_ALGO, "could not generate addr check");
1128 return;
1129 }
1130 /* ignore subq - not need for more init */
1131 }
1132 }
1133
1134 /**
1135 * Generate a NS check request to obtain authoritative information
1136 * on an NS rrset.
1137 *
1138 * @param qstate: the qtstate that triggered the need to prime.
1139 * @param iq: iterator query state.
1140 * @param id: module id.
1141 */
1142 static void
generate_ns_check(struct module_qstate * qstate,struct iter_qstate * iq,int id)1143 generate_ns_check(struct module_qstate* qstate, struct iter_qstate* iq, int id)
1144 {
1145 struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
1146 struct module_qstate* subq;
1147 log_assert(iq->dp);
1148
1149 if(iq->depth == ie->max_dependency_depth)
1150 return;
1151 if(!can_have_last_resort(qstate->env, iq->dp->name, iq->dp->namelen,
1152 iq->qchase.qclass, NULL))
1153 return;
1154 /* is this query the same as the nscheck? */
1155 if(qstate->qinfo.qtype == LDNS_RR_TYPE_NS &&
1156 query_dname_compare(iq->dp->name, qstate->qinfo.qname)==0 &&
1157 (qstate->query_flags&BIT_RD) && !(qstate->query_flags&BIT_CD)){
1158 /* spawn off A, AAAA queries for in-zone glue to check */
1159 generate_a_aaaa_check(qstate, iq, id);
1160 return;
1161 }
1162 /* no need to get the NS record for DS, it is above the zonecut */
1163 if(qstate->qinfo.qtype == LDNS_RR_TYPE_DS)
1164 return;
1165
1166 log_nametypeclass(VERB_ALGO, "schedule ns fetch",
1167 iq->dp->name, LDNS_RR_TYPE_NS, iq->qchase.qclass);
1168 if(!generate_sub_request(iq->dp->name, iq->dp->namelen,
1169 LDNS_RR_TYPE_NS, iq->qchase.qclass, qstate, id, iq,
1170 INIT_REQUEST_STATE, FINISHED_STATE, &subq, 1, 0)) {
1171 verbose(VERB_ALGO, "could not generate ns check");
1172 return;
1173 }
1174 if(subq) {
1175 struct iter_qstate* subiq =
1176 (struct iter_qstate*)subq->minfo[id];
1177
1178 /* make copy to avoid use of stub dp by different qs/threads */
1179 /* refetch glue to start higher up the tree */
1180 subiq->refetch_glue = 1;
1181 subiq->dp = delegpt_copy(iq->dp, subq->region);
1182 if(!subiq->dp) {
1183 log_err("out of memory generating ns check, copydp");
1184 fptr_ok(fptr_whitelist_modenv_kill_sub(
1185 qstate->env->kill_sub));
1186 (*qstate->env->kill_sub)(subq);
1187 return;
1188 }
1189 }
1190 }
1191
1192 /**
1193 * Generate a DNSKEY prefetch query to get the DNSKEY for the DS record we
1194 * just got in a referral (where we have dnssec_expected, thus have trust
1195 * anchors above it). Note that right after calling this routine the
1196 * iterator detached subqueries (because of following the referral), and thus
1197 * the DNSKEY query becomes detached, its return stored in the cache for
1198 * later lookup by the validator. This cache lookup by the validator avoids
1199 * the roundtrip incurred by the DNSKEY query. The DNSKEY query is now
1200 * performed at about the same time the original query is sent to the domain,
1201 * thus the two answers are likely to be returned at about the same time,
1202 * saving a roundtrip from the validated lookup.
1203 *
1204 * @param qstate: the qtstate that triggered the need to prime.
1205 * @param iq: iterator query state.
1206 * @param id: module id.
1207 */
1208 static void
generate_dnskey_prefetch(struct module_qstate * qstate,struct iter_qstate * iq,int id)1209 generate_dnskey_prefetch(struct module_qstate* qstate,
1210 struct iter_qstate* iq, int id)
1211 {
1212 struct module_qstate* subq;
1213 log_assert(iq->dp);
1214
1215 /* is this query the same as the prefetch? */
1216 if(qstate->qinfo.qtype == LDNS_RR_TYPE_DNSKEY &&
1217 query_dname_compare(iq->dp->name, qstate->qinfo.qname)==0 &&
1218 (qstate->query_flags&BIT_RD) && !(qstate->query_flags&BIT_CD)){
1219 return;
1220 }
1221 /* we do not generate this prefetch when the query list is full,
1222 * the query is fetched, if needed, when the validator wants it.
1223 * At that time the validator waits for it, after spawning it.
1224 * This means there is one state that uses cpu and a socket, the
1225 * spawned while this one waits, and not several at the same time,
1226 * if we had created the lookup here. And this helps to keep
1227 * the total load down, but the query still succeeds to resolve. */
1228 if(mesh_jostle_exceeded(qstate->env->mesh))
1229 return;
1230
1231 /* if the DNSKEY is in the cache this lookup will stop quickly */
1232 log_nametypeclass(VERB_ALGO, "schedule dnskey prefetch",
1233 iq->dp->name, LDNS_RR_TYPE_DNSKEY, iq->qchase.qclass);
1234 if(!generate_sub_request(iq->dp->name, iq->dp->namelen,
1235 LDNS_RR_TYPE_DNSKEY, iq->qchase.qclass, qstate, id, iq,
1236 INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0, 0)) {
1237 /* we'll be slower, but it'll work */
1238 verbose(VERB_ALGO, "could not generate dnskey prefetch");
1239 return;
1240 }
1241 if(subq) {
1242 struct iter_qstate* subiq =
1243 (struct iter_qstate*)subq->minfo[id];
1244 /* this qstate has the right delegation for the dnskey lookup*/
1245 /* make copy to avoid use of stub dp by different qs/threads */
1246 subiq->dp = delegpt_copy(iq->dp, subq->region);
1247 /* if !subiq->dp, it'll start from the cache, no problem */
1248 }
1249 }
1250
1251 /**
1252 * See if the query needs forwarding.
1253 *
1254 * @param qstate: query state.
1255 * @param iq: iterator query state.
1256 * @return true if the request is forwarded, false if not.
1257 * If returns true but, iq->dp is NULL then a malloc failure occurred.
1258 */
1259 static int
forward_request(struct module_qstate * qstate,struct iter_qstate * iq)1260 forward_request(struct module_qstate* qstate, struct iter_qstate* iq)
1261 {
1262 struct delegpt* dp;
1263 uint8_t* delname = iq->qchase.qname;
1264 size_t delnamelen = iq->qchase.qname_len;
1265 if(iq->refetch_glue && iq->dp) {
1266 delname = iq->dp->name;
1267 delnamelen = iq->dp->namelen;
1268 }
1269 /* strip one label off of DS query to lookup higher for it */
1270 if( (iq->qchase.qtype == LDNS_RR_TYPE_DS || iq->refetch_glue)
1271 && !dname_is_root(iq->qchase.qname))
1272 dname_remove_label(&delname, &delnamelen);
1273 dp = forwards_lookup(qstate->env->fwds, delname, iq->qchase.qclass);
1274 if(!dp)
1275 return 0;
1276 /* send recursion desired to forward addr */
1277 iq->chase_flags |= BIT_RD;
1278 iq->dp = delegpt_copy(dp, qstate->region);
1279 /* iq->dp checked by caller */
1280 verbose(VERB_ALGO, "forwarding request");
1281 return 1;
1282 }
1283
1284 /**
1285 * Process the initial part of the request handling. This state roughly
1286 * corresponds to resolver algorithms steps 1 (find answer in cache) and 2
1287 * (find the best servers to ask).
1288 *
1289 * Note that all requests start here, and query restarts revisit this state.
1290 *
1291 * This state either generates: 1) a response, from cache or error, 2) a
1292 * priming event, or 3) forwards the request to the next state (init2,
1293 * generally).
1294 *
1295 * @param qstate: query state.
1296 * @param iq: iterator query state.
1297 * @param ie: iterator shared global environment.
1298 * @param id: module id.
1299 * @return true if the event needs more request processing immediately,
1300 * false if not.
1301 */
1302 static int
processInitRequest(struct module_qstate * qstate,struct iter_qstate * iq,struct iter_env * ie,int id)1303 processInitRequest(struct module_qstate* qstate, struct iter_qstate* iq,
1304 struct iter_env* ie, int id)
1305 {
1306 uint8_t* delname, *dpname=NULL;
1307 size_t delnamelen, dpnamelen=0;
1308 struct dns_msg* msg = NULL;
1309
1310 log_query_info(VERB_DETAIL, "resolving", &qstate->qinfo);
1311 /* check effort */
1312
1313 /* We enforce a maximum number of query restarts. This is primarily a
1314 * cheap way to prevent CNAME loops. */
1315 if(iq->query_restart_count > MAX_RESTART_COUNT) {
1316 verbose(VERB_QUERY, "request has exceeded the maximum number"
1317 " of query restarts with %d", iq->query_restart_count);
1318 errinf(qstate, "request has exceeded the maximum number "
1319 "restarts (eg. indirections)");
1320 if(iq->qchase.qname)
1321 errinf_dname(qstate, "stop at", iq->qchase.qname);
1322 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1323 }
1324
1325 /* We enforce a maximum recursion/dependency depth -- in general,
1326 * this is unnecessary for dependency loops (although it will
1327 * catch those), but it provides a sensible limit to the amount
1328 * of work required to answer a given query. */
1329 verbose(VERB_ALGO, "request has dependency depth of %d", iq->depth);
1330 if(iq->depth > ie->max_dependency_depth) {
1331 verbose(VERB_QUERY, "request has exceeded the maximum "
1332 "dependency depth with depth of %d", iq->depth);
1333 errinf(qstate, "request has exceeded the maximum dependency "
1334 "depth (eg. nameserver lookup recursion)");
1335 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1336 }
1337
1338 /* If the request is qclass=ANY, setup to generate each class */
1339 if(qstate->qinfo.qclass == LDNS_RR_CLASS_ANY) {
1340 iq->qchase.qclass = 0;
1341 return next_state(iq, COLLECT_CLASS_STATE);
1342 }
1343
1344 /*
1345 * If we are restricted by a forward-zone or a stub-zone, we
1346 * can't re-fetch glue for this delegation point.
1347 * we won’t try to re-fetch glue if the iq->dp is null.
1348 */
1349 if (iq->refetch_glue &&
1350 iq->dp &&
1351 !can_have_last_resort(qstate->env, iq->dp->name,
1352 iq->dp->namelen, iq->qchase.qclass, NULL)) {
1353 iq->refetch_glue = 0;
1354 }
1355
1356 /* Resolver Algorithm Step 1 -- Look for the answer in local data. */
1357
1358 /* This either results in a query restart (CNAME cache response), a
1359 * terminating response (ANSWER), or a cache miss (null). */
1360
1361 if (iter_stub_fwd_no_cache(qstate, &iq->qchase, &dpname, &dpnamelen)) {
1362 /* Asked to not query cache. */
1363 verbose(VERB_ALGO, "no-cache set, going to the network");
1364 qstate->no_cache_lookup = 1;
1365 qstate->no_cache_store = 1;
1366 msg = NULL;
1367 } else if(qstate->blacklist) {
1368 /* if cache, or anything else, was blacklisted then
1369 * getting older results from cache is a bad idea, no cache */
1370 verbose(VERB_ALGO, "cache blacklisted, going to the network");
1371 msg = NULL;
1372 } else if(!qstate->no_cache_lookup) {
1373 msg = dns_cache_lookup(qstate->env, iq->qchase.qname,
1374 iq->qchase.qname_len, iq->qchase.qtype,
1375 iq->qchase.qclass, qstate->query_flags,
1376 qstate->region, qstate->env->scratch, 0, dpname,
1377 dpnamelen);
1378 if(!msg && qstate->env->neg_cache &&
1379 iter_qname_indicates_dnssec(qstate->env, &iq->qchase)) {
1380 /* lookup in negative cache; may result in
1381 * NOERROR/NODATA or NXDOMAIN answers that need validation */
1382 msg = val_neg_getmsg(qstate->env->neg_cache, &iq->qchase,
1383 qstate->region, qstate->env->rrset_cache,
1384 qstate->env->scratch_buffer,
1385 *qstate->env->now, 1/*add SOA*/, NULL,
1386 qstate->env->cfg);
1387 }
1388 /* item taken from cache does not match our query name, thus
1389 * security needs to be re-examined later */
1390 if(msg && query_dname_compare(qstate->qinfo.qname,
1391 iq->qchase.qname) != 0)
1392 msg->rep->security = sec_status_unchecked;
1393 }
1394 if(msg) {
1395 /* handle positive cache response */
1396 enum response_type type = response_type_from_cache(msg,
1397 &iq->qchase);
1398 if(verbosity >= VERB_ALGO) {
1399 log_dns_msg("msg from cache lookup", &msg->qinfo,
1400 msg->rep);
1401 verbose(VERB_ALGO, "msg ttl is %d, prefetch ttl %d",
1402 (int)msg->rep->ttl,
1403 (int)msg->rep->prefetch_ttl);
1404 }
1405
1406 if(type == RESPONSE_TYPE_CNAME) {
1407 uint8_t* sname = 0;
1408 size_t slen = 0;
1409 verbose(VERB_ALGO, "returning CNAME response from "
1410 "cache");
1411 if(!handle_cname_response(qstate, iq, msg,
1412 &sname, &slen)) {
1413 errinf(qstate, "failed to prepend CNAME "
1414 "components, malloc failure");
1415 return error_response(qstate, id,
1416 LDNS_RCODE_SERVFAIL);
1417 }
1418 iq->qchase.qname = sname;
1419 iq->qchase.qname_len = slen;
1420 /* This *is* a query restart, even if it is a cheap
1421 * one. */
1422 iq->dp = NULL;
1423 iq->refetch_glue = 0;
1424 iq->query_restart_count++;
1425 iq->sent_count = 0;
1426 iq->dp_target_count = 0;
1427 sock_list_insert(&qstate->reply_origin, NULL, 0, qstate->region);
1428 if(qstate->env->cfg->qname_minimisation)
1429 iq->minimisation_state = INIT_MINIMISE_STATE;
1430 return next_state(iq, INIT_REQUEST_STATE);
1431 }
1432
1433 /* if from cache, NULL, else insert 'cache IP' len=0 */
1434 if(qstate->reply_origin)
1435 sock_list_insert(&qstate->reply_origin, NULL, 0, qstate->region);
1436 if(FLAGS_GET_RCODE(msg->rep->flags) == LDNS_RCODE_SERVFAIL)
1437 errinf(qstate, "SERVFAIL in cache");
1438 /* it is an answer, response, to final state */
1439 verbose(VERB_ALGO, "returning answer from cache.");
1440 iq->response = msg;
1441 return final_state(iq);
1442 }
1443
1444 /* attempt to forward the request */
1445 if(forward_request(qstate, iq))
1446 {
1447 if(!iq->dp) {
1448 log_err("alloc failure for forward dp");
1449 errinf(qstate, "malloc failure for forward zone");
1450 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1451 }
1452 iq->refetch_glue = 0;
1453 iq->minimisation_state = DONOT_MINIMISE_STATE;
1454 /* the request has been forwarded.
1455 * forwarded requests need to be immediately sent to the
1456 * next state, QUERYTARGETS. */
1457 return next_state(iq, QUERYTARGETS_STATE);
1458 }
1459
1460 /* Resolver Algorithm Step 2 -- find the "best" servers. */
1461
1462 /* first, adjust for DS queries. To avoid the grandparent problem,
1463 * we just look for the closest set of server to the parent of qname.
1464 * When re-fetching glue we also need to ask the parent.
1465 */
1466 if(iq->refetch_glue) {
1467 if(!iq->dp) {
1468 log_err("internal or malloc fail: no dp for refetch");
1469 errinf(qstate, "malloc failure, for delegation info");
1470 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1471 }
1472 delname = iq->dp->name;
1473 delnamelen = iq->dp->namelen;
1474 } else {
1475 delname = iq->qchase.qname;
1476 delnamelen = iq->qchase.qname_len;
1477 }
1478 if(iq->qchase.qtype == LDNS_RR_TYPE_DS || iq->refetch_glue ||
1479 (iq->qchase.qtype == LDNS_RR_TYPE_NS && qstate->prefetch_leeway
1480 && can_have_last_resort(qstate->env, delname, delnamelen, iq->qchase.qclass, NULL))) {
1481 /* remove first label from delname, root goes to hints,
1482 * but only to fetch glue, not for qtype=DS. */
1483 /* also when prefetching an NS record, fetch it again from
1484 * its parent, just as if it expired, so that you do not
1485 * get stuck on an older nameserver that gives old NSrecords */
1486 if(dname_is_root(delname) && (iq->refetch_glue ||
1487 (iq->qchase.qtype == LDNS_RR_TYPE_NS &&
1488 qstate->prefetch_leeway)))
1489 delname = NULL; /* go to root priming */
1490 else dname_remove_label(&delname, &delnamelen);
1491 }
1492 /* delname is the name to lookup a delegation for. If NULL rootprime */
1493 while(1) {
1494
1495 /* Lookup the delegation in the cache. If null, then the
1496 * cache needs to be primed for the qclass. */
1497 if(delname)
1498 iq->dp = dns_cache_find_delegation(qstate->env, delname,
1499 delnamelen, iq->qchase.qtype, iq->qchase.qclass,
1500 qstate->region, &iq->deleg_msg,
1501 *qstate->env->now+qstate->prefetch_leeway, 1,
1502 dpname, dpnamelen);
1503 else iq->dp = NULL;
1504
1505 /* If the cache has returned nothing, then we have a
1506 * root priming situation. */
1507 if(iq->dp == NULL) {
1508 int r;
1509 /* if under auth zone, no prime needed */
1510 if(!auth_zone_delegpt(qstate, iq, delname, delnamelen))
1511 return error_response(qstate, id,
1512 LDNS_RCODE_SERVFAIL);
1513 if(iq->dp) /* use auth zone dp */
1514 return next_state(iq, INIT_REQUEST_2_STATE);
1515 /* if there is a stub, then no root prime needed */
1516 r = prime_stub(qstate, iq, id, delname,
1517 iq->qchase.qclass);
1518 if(r == 2)
1519 break; /* got noprime-stub-zone, continue */
1520 else if(r)
1521 return 0; /* stub prime request made */
1522 if(forwards_lookup_root(qstate->env->fwds,
1523 iq->qchase.qclass)) {
1524 /* forward zone root, no root prime needed */
1525 /* fill in some dp - safety belt */
1526 iq->dp = hints_lookup_root(qstate->env->hints,
1527 iq->qchase.qclass);
1528 if(!iq->dp) {
1529 log_err("internal error: no hints dp");
1530 errinf(qstate, "no hints for this class");
1531 return error_response(qstate, id,
1532 LDNS_RCODE_SERVFAIL);
1533 }
1534 iq->dp = delegpt_copy(iq->dp, qstate->region);
1535 if(!iq->dp) {
1536 log_err("out of memory in safety belt");
1537 errinf(qstate, "malloc failure, in safety belt");
1538 return error_response(qstate, id,
1539 LDNS_RCODE_SERVFAIL);
1540 }
1541 return next_state(iq, INIT_REQUEST_2_STATE);
1542 }
1543 /* Note that the result of this will set a new
1544 * DelegationPoint based on the result of priming. */
1545 if(!prime_root(qstate, iq, id, iq->qchase.qclass))
1546 return error_response(qstate, id,
1547 LDNS_RCODE_REFUSED);
1548
1549 /* priming creates and sends a subordinate query, with
1550 * this query as the parent. So further processing for
1551 * this event will stop until reactivated by the
1552 * results of priming. */
1553 return 0;
1554 }
1555 if(!iq->ratelimit_ok && qstate->prefetch_leeway)
1556 iq->ratelimit_ok = 1; /* allow prefetches, this keeps
1557 otherwise valid data in the cache */
1558
1559 /* see if this dp not useless.
1560 * It is useless if:
1561 * o all NS items are required glue.
1562 * or the query is for NS item that is required glue.
1563 * o no addresses are provided.
1564 * o RD qflag is on.
1565 * Instead, go up one level, and try to get even further
1566 * If the root was useless, use safety belt information.
1567 * Only check cache returns, because replies for servers
1568 * could be useless but lead to loops (bumping into the
1569 * same server reply) if useless-checked.
1570 */
1571 if(iter_dp_is_useless(&qstate->qinfo, qstate->query_flags,
1572 iq->dp, ie->supports_ipv4, ie->supports_ipv6)) {
1573 struct delegpt* retdp = NULL;
1574 if(!can_have_last_resort(qstate->env, iq->dp->name, iq->dp->namelen, iq->qchase.qclass, &retdp)) {
1575 if(retdp) {
1576 verbose(VERB_QUERY, "cache has stub "
1577 "or fwd but no addresses, "
1578 "fallback to config");
1579 iq->dp = delegpt_copy(retdp,
1580 qstate->region);
1581 if(!iq->dp) {
1582 log_err("out of memory in "
1583 "stub/fwd fallback");
1584 errinf(qstate, "malloc failure, for fallback to config");
1585 return error_response(qstate,
1586 id, LDNS_RCODE_SERVFAIL);
1587 }
1588 break;
1589 }
1590 verbose(VERB_ALGO, "useless dp "
1591 "but cannot go up, servfail");
1592 delegpt_log(VERB_ALGO, iq->dp);
1593 errinf(qstate, "no useful nameservers, "
1594 "and cannot go up");
1595 errinf_dname(qstate, "for zone", iq->dp->name);
1596 return error_response(qstate, id,
1597 LDNS_RCODE_SERVFAIL);
1598 }
1599 if(dname_is_root(iq->dp->name)) {
1600 /* use safety belt */
1601 verbose(VERB_QUERY, "Cache has root NS but "
1602 "no addresses. Fallback to the safety belt.");
1603 iq->dp = hints_lookup_root(qstate->env->hints,
1604 iq->qchase.qclass);
1605 /* note deleg_msg is from previous lookup,
1606 * but RD is on, so it is not used */
1607 if(!iq->dp) {
1608 log_err("internal error: no hints dp");
1609 return error_response(qstate, id,
1610 LDNS_RCODE_REFUSED);
1611 }
1612 iq->dp = delegpt_copy(iq->dp, qstate->region);
1613 if(!iq->dp) {
1614 log_err("out of memory in safety belt");
1615 errinf(qstate, "malloc failure, in safety belt, for root");
1616 return error_response(qstate, id,
1617 LDNS_RCODE_SERVFAIL);
1618 }
1619 break;
1620 } else {
1621 verbose(VERB_ALGO,
1622 "cache delegation was useless:");
1623 delegpt_log(VERB_ALGO, iq->dp);
1624 /* go up */
1625 delname = iq->dp->name;
1626 delnamelen = iq->dp->namelen;
1627 dname_remove_label(&delname, &delnamelen);
1628 }
1629 } else break;
1630 }
1631
1632 verbose(VERB_ALGO, "cache delegation returns delegpt");
1633 delegpt_log(VERB_ALGO, iq->dp);
1634
1635 /* Otherwise, set the current delegation point and move on to the
1636 * next state. */
1637 return next_state(iq, INIT_REQUEST_2_STATE);
1638 }
1639
1640 /**
1641 * Process the second part of the initial request handling. This state
1642 * basically exists so that queries that generate root priming events have
1643 * the same init processing as ones that do not. Request events that reach
1644 * this state must have a valid currentDelegationPoint set.
1645 *
1646 * This part is primarily handling stub zone priming. Events that reach this
1647 * state must have a current delegation point.
1648 *
1649 * @param qstate: query state.
1650 * @param iq: iterator query state.
1651 * @param id: module id.
1652 * @return true if the event needs more request processing immediately,
1653 * false if not.
1654 */
1655 static int
processInitRequest2(struct module_qstate * qstate,struct iter_qstate * iq,int id)1656 processInitRequest2(struct module_qstate* qstate, struct iter_qstate* iq,
1657 int id)
1658 {
1659 uint8_t* delname;
1660 size_t delnamelen;
1661 log_query_info(VERB_QUERY, "resolving (init part 2): ",
1662 &qstate->qinfo);
1663
1664 delname = iq->qchase.qname;
1665 delnamelen = iq->qchase.qname_len;
1666 if(iq->refetch_glue) {
1667 struct iter_hints_stub* stub;
1668 if(!iq->dp) {
1669 log_err("internal or malloc fail: no dp for refetch");
1670 errinf(qstate, "malloc failure, no delegation info");
1671 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1672 }
1673 /* Do not send queries above stub, do not set delname to dp if
1674 * this is above stub without stub-first. */
1675 stub = hints_lookup_stub(
1676 qstate->env->hints, iq->qchase.qname, iq->qchase.qclass,
1677 iq->dp);
1678 if(!stub || !stub->dp->has_parent_side_NS ||
1679 dname_subdomain_c(iq->dp->name, stub->dp->name)) {
1680 delname = iq->dp->name;
1681 delnamelen = iq->dp->namelen;
1682 }
1683 }
1684 if(iq->qchase.qtype == LDNS_RR_TYPE_DS || iq->refetch_glue) {
1685 if(!dname_is_root(delname))
1686 dname_remove_label(&delname, &delnamelen);
1687 iq->refetch_glue = 0; /* if CNAME causes restart, no refetch */
1688 }
1689
1690 /* see if we have an auth zone to answer from, improves dp from cache
1691 * (if any dp from cache) with auth zone dp, if that is lower */
1692 if(!auth_zone_delegpt(qstate, iq, delname, delnamelen))
1693 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1694
1695 /* Check to see if we need to prime a stub zone. */
1696 if(prime_stub(qstate, iq, id, delname, iq->qchase.qclass)) {
1697 /* A priming sub request was made */
1698 return 0;
1699 }
1700
1701 /* most events just get forwarded to the next state. */
1702 return next_state(iq, INIT_REQUEST_3_STATE);
1703 }
1704
1705 /**
1706 * Process the third part of the initial request handling. This state exists
1707 * as a separate state so that queries that generate stub priming events
1708 * will get the tail end of the init process but not repeat the stub priming
1709 * check.
1710 *
1711 * @param qstate: query state.
1712 * @param iq: iterator query state.
1713 * @param id: module id.
1714 * @return true, advancing the event to the QUERYTARGETS_STATE.
1715 */
1716 static int
processInitRequest3(struct module_qstate * qstate,struct iter_qstate * iq,int id)1717 processInitRequest3(struct module_qstate* qstate, struct iter_qstate* iq,
1718 int id)
1719 {
1720 log_query_info(VERB_QUERY, "resolving (init part 3): ",
1721 &qstate->qinfo);
1722 /* if the cache reply dp equals a validation anchor or msg has DS,
1723 * then DNSSEC RRSIGs are expected in the reply */
1724 iq->dnssec_expected = iter_indicates_dnssec(qstate->env, iq->dp,
1725 iq->deleg_msg, iq->qchase.qclass);
1726
1727 /* If the RD flag wasn't set, then we just finish with the
1728 * cached referral as the response. */
1729 if(!(qstate->query_flags & BIT_RD) && iq->deleg_msg) {
1730 iq->response = iq->deleg_msg;
1731 if(verbosity >= VERB_ALGO && iq->response)
1732 log_dns_msg("no RD requested, using delegation msg",
1733 &iq->response->qinfo, iq->response->rep);
1734 if(qstate->reply_origin)
1735 sock_list_insert(&qstate->reply_origin, NULL, 0, qstate->region);
1736 return final_state(iq);
1737 }
1738 /* After this point, unset the RD flag -- this query is going to
1739 * be sent to an auth. server. */
1740 iq->chase_flags &= ~BIT_RD;
1741
1742 /* if dnssec expected, fetch key for the trust-anchor or cached-DS */
1743 if(iq->dnssec_expected && qstate->env->cfg->prefetch_key &&
1744 !(qstate->query_flags&BIT_CD)) {
1745 generate_dnskey_prefetch(qstate, iq, id);
1746 fptr_ok(fptr_whitelist_modenv_detach_subs(
1747 qstate->env->detach_subs));
1748 (*qstate->env->detach_subs)(qstate);
1749 }
1750
1751 /* Jump to the next state. */
1752 return next_state(iq, QUERYTARGETS_STATE);
1753 }
1754
1755 /**
1756 * Given a basic query, generate a parent-side "target" query.
1757 * These are subordinate queries for missing delegation point target addresses,
1758 * for which only the parent of the delegation provides correct IP addresses.
1759 *
1760 * @param qstate: query state.
1761 * @param iq: iterator query state.
1762 * @param id: module id.
1763 * @param name: target qname.
1764 * @param namelen: target qname length.
1765 * @param qtype: target qtype (either A or AAAA).
1766 * @param qclass: target qclass.
1767 * @return true on success, false on failure.
1768 */
1769 static int
generate_parentside_target_query(struct module_qstate * qstate,struct iter_qstate * iq,int id,uint8_t * name,size_t namelen,uint16_t qtype,uint16_t qclass)1770 generate_parentside_target_query(struct module_qstate* qstate,
1771 struct iter_qstate* iq, int id, uint8_t* name, size_t namelen,
1772 uint16_t qtype, uint16_t qclass)
1773 {
1774 struct module_qstate* subq;
1775 if(!generate_sub_request(name, namelen, qtype, qclass, qstate,
1776 id, iq, INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0, 0))
1777 return 0;
1778 if(subq) {
1779 struct iter_qstate* subiq =
1780 (struct iter_qstate*)subq->minfo[id];
1781 /* blacklist the cache - we want to fetch parent stuff */
1782 sock_list_insert(&subq->blacklist, NULL, 0, subq->region);
1783 subiq->query_for_pside_glue = 1;
1784 if(dname_subdomain_c(name, iq->dp->name)) {
1785 subiq->dp = delegpt_copy(iq->dp, subq->region);
1786 subiq->dnssec_expected = iter_indicates_dnssec(
1787 qstate->env, subiq->dp, NULL,
1788 subq->qinfo.qclass);
1789 subiq->refetch_glue = 1;
1790 } else {
1791 subiq->dp = dns_cache_find_delegation(qstate->env,
1792 name, namelen, qtype, qclass, subq->region,
1793 &subiq->deleg_msg,
1794 *qstate->env->now+subq->prefetch_leeway,
1795 1, NULL, 0);
1796 /* if no dp, then it's from root, refetch unneeded */
1797 if(subiq->dp) {
1798 subiq->dnssec_expected = iter_indicates_dnssec(
1799 qstate->env, subiq->dp, NULL,
1800 subq->qinfo.qclass);
1801 subiq->refetch_glue = 1;
1802 }
1803 }
1804 }
1805 log_nametypeclass(VERB_QUERY, "new pside target", name, qtype, qclass);
1806 return 1;
1807 }
1808
1809 /**
1810 * Given a basic query, generate a "target" query. These are subordinate
1811 * queries for missing delegation point target addresses.
1812 *
1813 * @param qstate: query state.
1814 * @param iq: iterator query state.
1815 * @param id: module id.
1816 * @param name: target qname.
1817 * @param namelen: target qname length.
1818 * @param qtype: target qtype (either A or AAAA).
1819 * @param qclass: target qclass.
1820 * @return true on success, false on failure.
1821 */
1822 static int
generate_target_query(struct module_qstate * qstate,struct iter_qstate * iq,int id,uint8_t * name,size_t namelen,uint16_t qtype,uint16_t qclass)1823 generate_target_query(struct module_qstate* qstate, struct iter_qstate* iq,
1824 int id, uint8_t* name, size_t namelen, uint16_t qtype, uint16_t qclass)
1825 {
1826 struct module_qstate* subq;
1827 if(!generate_sub_request(name, namelen, qtype, qclass, qstate,
1828 id, iq, INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0, 0))
1829 return 0;
1830 log_nametypeclass(VERB_QUERY, "new target", name, qtype, qclass);
1831 return 1;
1832 }
1833
1834 /**
1835 * Given an event at a certain state, generate zero or more target queries
1836 * for it's current delegation point.
1837 *
1838 * @param qstate: query state.
1839 * @param iq: iterator query state.
1840 * @param ie: iterator shared global environment.
1841 * @param id: module id.
1842 * @param maxtargets: The maximum number of targets to query for.
1843 * if it is negative, there is no maximum number of targets.
1844 * @param num: returns the number of queries generated and processed,
1845 * which may be zero if there were no missing targets.
1846 * @return false on error.
1847 */
1848 static int
query_for_targets(struct module_qstate * qstate,struct iter_qstate * iq,struct iter_env * ie,int id,int maxtargets,int * num)1849 query_for_targets(struct module_qstate* qstate, struct iter_qstate* iq,
1850 struct iter_env* ie, int id, int maxtargets, int* num)
1851 {
1852 int query_count = 0;
1853 struct delegpt_ns* ns;
1854 int missing;
1855 int toget = 0;
1856
1857 iter_mark_cycle_targets(qstate, iq->dp);
1858 missing = (int)delegpt_count_missing_targets(iq->dp, NULL);
1859 log_assert(maxtargets != 0); /* that would not be useful */
1860
1861 /* Generate target requests. Basically, any missing targets
1862 * are queried for here, regardless if it is necessary to do
1863 * so to continue processing. */
1864 if(maxtargets < 0 || maxtargets > missing)
1865 toget = missing;
1866 else toget = maxtargets;
1867 if(toget == 0) {
1868 *num = 0;
1869 return 1;
1870 }
1871
1872 /* now that we are sure that a target query is going to be made,
1873 * check the limits. */
1874 if(iq->depth == ie->max_dependency_depth)
1875 return 0;
1876 if(iq->depth > 0 && iq->target_count &&
1877 iq->target_count[TARGET_COUNT_QUERIES] > MAX_TARGET_COUNT) {
1878 char s[LDNS_MAX_DOMAINLEN+1];
1879 dname_str(qstate->qinfo.qname, s);
1880 verbose(VERB_QUERY, "request %s has exceeded the maximum "
1881 "number of glue fetches %d", s,
1882 iq->target_count[TARGET_COUNT_QUERIES]);
1883 return 0;
1884 }
1885 if(iq->dp_target_count > MAX_DP_TARGET_COUNT) {
1886 char s[LDNS_MAX_DOMAINLEN+1];
1887 dname_str(qstate->qinfo.qname, s);
1888 verbose(VERB_QUERY, "request %s has exceeded the maximum "
1889 "number of glue fetches %d to a single delegation point",
1890 s, iq->dp_target_count);
1891 return 0;
1892 }
1893
1894 /* select 'toget' items from the total of 'missing' items */
1895 log_assert(toget <= missing);
1896
1897 /* loop over missing targets */
1898 for(ns = iq->dp->nslist; ns; ns = ns->next) {
1899 if(ns->resolved)
1900 continue;
1901
1902 /* randomly select this item with probability toget/missing */
1903 if(!iter_ns_probability(qstate->env->rnd, toget, missing)) {
1904 /* do not select this one, next; select toget number
1905 * of items from a list one less in size */
1906 missing --;
1907 continue;
1908 }
1909
1910 if(ie->supports_ipv6 &&
1911 ((ns->lame && !ns->done_pside6) ||
1912 (!ns->lame && !ns->got6))) {
1913 /* Send the AAAA request. */
1914 if(!generate_target_query(qstate, iq, id,
1915 ns->name, ns->namelen,
1916 LDNS_RR_TYPE_AAAA, iq->qchase.qclass)) {
1917 *num = query_count;
1918 if(query_count > 0)
1919 qstate->ext_state[id] = module_wait_subquery;
1920 return 0;
1921 }
1922 query_count++;
1923 /* If the mesh query list is full, exit the loop here.
1924 * This makes the routine spawn one query at a time,
1925 * and this means there is no query state load
1926 * increase, because the spawned state uses cpu and a
1927 * socket while this state waits for that spawned
1928 * state. Next time we can look up further targets */
1929 if(mesh_jostle_exceeded(qstate->env->mesh))
1930 break;
1931 }
1932 /* Send the A request. */
1933 if(ie->supports_ipv4 &&
1934 ((ns->lame && !ns->done_pside4) ||
1935 (!ns->lame && !ns->got4))) {
1936 if(!generate_target_query(qstate, iq, id,
1937 ns->name, ns->namelen,
1938 LDNS_RR_TYPE_A, iq->qchase.qclass)) {
1939 *num = query_count;
1940 if(query_count > 0)
1941 qstate->ext_state[id] = module_wait_subquery;
1942 return 0;
1943 }
1944 query_count++;
1945 /* If the mesh query list is full, exit the loop. */
1946 if(mesh_jostle_exceeded(qstate->env->mesh))
1947 break;
1948 }
1949
1950 /* mark this target as in progress. */
1951 ns->resolved = 1;
1952 missing--;
1953 toget--;
1954 if(toget == 0)
1955 break;
1956 }
1957 *num = query_count;
1958 if(query_count > 0)
1959 qstate->ext_state[id] = module_wait_subquery;
1960
1961 return 1;
1962 }
1963
1964 /**
1965 * Called by processQueryTargets when it would like extra targets to query
1966 * but it seems to be out of options. At last resort some less appealing
1967 * options are explored. If there are no more options, the result is SERVFAIL
1968 *
1969 * @param qstate: query state.
1970 * @param iq: iterator query state.
1971 * @param ie: iterator shared global environment.
1972 * @param id: module id.
1973 * @return true if the event requires more request processing immediately,
1974 * false if not.
1975 */
1976 static int
processLastResort(struct module_qstate * qstate,struct iter_qstate * iq,struct iter_env * ie,int id)1977 processLastResort(struct module_qstate* qstate, struct iter_qstate* iq,
1978 struct iter_env* ie, int id)
1979 {
1980 struct delegpt_ns* ns;
1981 int query_count = 0;
1982 verbose(VERB_ALGO, "No more query targets, attempting last resort");
1983 log_assert(iq->dp);
1984
1985 if(!can_have_last_resort(qstate->env, iq->dp->name, iq->dp->namelen,
1986 iq->qchase.qclass, NULL)) {
1987 /* fail -- no more targets, no more hope of targets, no hope
1988 * of a response. */
1989 errinf(qstate, "all the configured stub or forward servers failed,");
1990 errinf_dname(qstate, "at zone", iq->dp->name);
1991 errinf_reply(qstate, iq);
1992 verbose(VERB_QUERY, "configured stub or forward servers failed -- returning SERVFAIL");
1993 return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
1994 }
1995 if(!iq->dp->has_parent_side_NS && dname_is_root(iq->dp->name)) {
1996 struct delegpt* p = hints_lookup_root(qstate->env->hints,
1997 iq->qchase.qclass);
1998 if(p) {
1999 struct delegpt_addr* a;
2000 iq->chase_flags &= ~BIT_RD; /* go to authorities */
2001 for(ns = p->nslist; ns; ns=ns->next) {
2002 (void)delegpt_add_ns(iq->dp, qstate->region,
2003 ns->name, ns->lame, ns->tls_auth_name,
2004 ns->port);
2005 }
2006 for(a = p->target_list; a; a=a->next_target) {
2007 (void)delegpt_add_addr(iq->dp, qstate->region,
2008 &a->addr, a->addrlen, a->bogus,
2009 a->lame, a->tls_auth_name, -1, NULL);
2010 }
2011 }
2012 iq->dp->has_parent_side_NS = 1;
2013 } else if(!iq->dp->has_parent_side_NS) {
2014 if(!iter_lookup_parent_NS_from_cache(qstate->env, iq->dp,
2015 qstate->region, &qstate->qinfo)
2016 || !iq->dp->has_parent_side_NS) {
2017 /* if: malloc failure in lookup go up to try */
2018 /* if: no parent NS in cache - go up one level */
2019 verbose(VERB_ALGO, "try to grab parent NS");
2020 iq->store_parent_NS = iq->dp;
2021 iq->chase_flags &= ~BIT_RD; /* go to authorities */
2022 iq->deleg_msg = NULL;
2023 iq->refetch_glue = 1;
2024 iq->query_restart_count++;
2025 iq->sent_count = 0;
2026 iq->dp_target_count = 0;
2027 if(qstate->env->cfg->qname_minimisation)
2028 iq->minimisation_state = INIT_MINIMISE_STATE;
2029 return next_state(iq, INIT_REQUEST_STATE);
2030 }
2031 }
2032 /* see if that makes new names available */
2033 if(!cache_fill_missing(qstate->env, iq->qchase.qclass,
2034 qstate->region, iq->dp))
2035 log_err("out of memory in cache_fill_missing");
2036 if(iq->dp->usable_list) {
2037 verbose(VERB_ALGO, "try parent-side-name, w. glue from cache");
2038 return next_state(iq, QUERYTARGETS_STATE);
2039 }
2040 /* try to fill out parent glue from cache */
2041 if(iter_lookup_parent_glue_from_cache(qstate->env, iq->dp,
2042 qstate->region, &qstate->qinfo)) {
2043 /* got parent stuff from cache, see if we can continue */
2044 verbose(VERB_ALGO, "try parent-side glue from cache");
2045 return next_state(iq, QUERYTARGETS_STATE);
2046 }
2047 /* query for an extra name added by the parent-NS record */
2048 if(delegpt_count_missing_targets(iq->dp, NULL) > 0) {
2049 int qs = 0;
2050 verbose(VERB_ALGO, "try parent-side target name");
2051 if(!query_for_targets(qstate, iq, ie, id, 1, &qs)) {
2052 errinf(qstate, "could not fetch nameserver");
2053 errinf_dname(qstate, "at zone", iq->dp->name);
2054 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2055 }
2056 iq->num_target_queries += qs;
2057 target_count_increase(iq, qs);
2058 if(qs != 0) {
2059 qstate->ext_state[id] = module_wait_subquery;
2060 return 0; /* and wait for them */
2061 }
2062 }
2063 if(iq->depth == ie->max_dependency_depth) {
2064 verbose(VERB_QUERY, "maxdepth and need more nameservers, fail");
2065 errinf(qstate, "cannot fetch more nameservers because at max dependency depth");
2066 return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
2067 }
2068 if(iq->depth > 0 && iq->target_count &&
2069 iq->target_count[TARGET_COUNT_QUERIES] > MAX_TARGET_COUNT) {
2070 char s[LDNS_MAX_DOMAINLEN+1];
2071 dname_str(qstate->qinfo.qname, s);
2072 verbose(VERB_QUERY, "request %s has exceeded the maximum "
2073 "number of glue fetches %d", s,
2074 iq->target_count[TARGET_COUNT_QUERIES]);
2075 errinf(qstate, "exceeded the maximum number of glue fetches");
2076 return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
2077 }
2078 /* mark cycle targets for parent-side lookups */
2079 iter_mark_pside_cycle_targets(qstate, iq->dp);
2080 /* see if we can issue queries to get nameserver addresses */
2081 /* this lookup is not randomized, but sequential. */
2082 for(ns = iq->dp->nslist; ns; ns = ns->next) {
2083 /* if this nameserver is at a delegation point, but that
2084 * delegation point is a stub and we cannot go higher, skip*/
2085 if( ((ie->supports_ipv6 && !ns->done_pside6) ||
2086 (ie->supports_ipv4 && !ns->done_pside4)) &&
2087 !can_have_last_resort(qstate->env, ns->name, ns->namelen,
2088 iq->qchase.qclass, NULL)) {
2089 log_nametypeclass(VERB_ALGO, "cannot pside lookup ns "
2090 "because it is also a stub/forward,",
2091 ns->name, LDNS_RR_TYPE_NS, iq->qchase.qclass);
2092 if(ie->supports_ipv6) ns->done_pside6 = 1;
2093 if(ie->supports_ipv4) ns->done_pside4 = 1;
2094 continue;
2095 }
2096 /* query for parent-side A and AAAA for nameservers */
2097 if(ie->supports_ipv6 && !ns->done_pside6) {
2098 /* Send the AAAA request. */
2099 if(!generate_parentside_target_query(qstate, iq, id,
2100 ns->name, ns->namelen,
2101 LDNS_RR_TYPE_AAAA, iq->qchase.qclass)) {
2102 errinf_dname(qstate, "could not generate nameserver AAAA lookup for", ns->name);
2103 return error_response(qstate, id,
2104 LDNS_RCODE_SERVFAIL);
2105 }
2106 ns->done_pside6 = 1;
2107 query_count++;
2108 if(mesh_jostle_exceeded(qstate->env->mesh)) {
2109 /* Wait for the lookup; do not spawn multiple
2110 * lookups at a time. */
2111 verbose(VERB_ALGO, "try parent-side glue lookup");
2112 iq->num_target_queries += query_count;
2113 target_count_increase(iq, query_count);
2114 qstate->ext_state[id] = module_wait_subquery;
2115 return 0;
2116 }
2117 }
2118 if(ie->supports_ipv4 && !ns->done_pside4) {
2119 /* Send the A request. */
2120 if(!generate_parentside_target_query(qstate, iq, id,
2121 ns->name, ns->namelen,
2122 LDNS_RR_TYPE_A, iq->qchase.qclass)) {
2123 errinf_dname(qstate, "could not generate nameserver A lookup for", ns->name);
2124 return error_response(qstate, id,
2125 LDNS_RCODE_SERVFAIL);
2126 }
2127 ns->done_pside4 = 1;
2128 query_count++;
2129 }
2130 if(query_count != 0) { /* suspend to await results */
2131 verbose(VERB_ALGO, "try parent-side glue lookup");
2132 iq->num_target_queries += query_count;
2133 target_count_increase(iq, query_count);
2134 qstate->ext_state[id] = module_wait_subquery;
2135 return 0;
2136 }
2137 }
2138
2139 /* if this was a parent-side glue query itself, then store that
2140 * failure in cache. */
2141 if(!qstate->no_cache_store && iq->query_for_pside_glue
2142 && !iq->pside_glue)
2143 iter_store_parentside_neg(qstate->env, &qstate->qinfo,
2144 iq->deleg_msg?iq->deleg_msg->rep:
2145 (iq->response?iq->response->rep:NULL));
2146
2147 errinf(qstate, "all servers for this domain failed,");
2148 errinf_dname(qstate, "at zone", iq->dp->name);
2149 errinf_reply(qstate, iq);
2150 verbose(VERB_QUERY, "out of query targets -- returning SERVFAIL");
2151 /* fail -- no more targets, no more hope of targets, no hope
2152 * of a response. */
2153 return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
2154 }
2155
2156 /**
2157 * Try to find the NS record set that will resolve a qtype DS query. Due
2158 * to grandparent/grandchild reasons we did not get a proper lookup right
2159 * away. We need to create type NS queries until we get the right parent
2160 * for this lookup. We remove labels from the query to find the right point.
2161 * If we end up at the old dp name, then there is no solution.
2162 *
2163 * @param qstate: query state.
2164 * @param iq: iterator query state.
2165 * @param id: module id.
2166 * @return true if the event requires more immediate processing, false if
2167 * not. This is generally only true when forwarding the request to
2168 * the final state (i.e., on answer).
2169 */
2170 static int
processDSNSFind(struct module_qstate * qstate,struct iter_qstate * iq,int id)2171 processDSNSFind(struct module_qstate* qstate, struct iter_qstate* iq, int id)
2172 {
2173 struct module_qstate* subq = NULL;
2174 verbose(VERB_ALGO, "processDSNSFind");
2175
2176 if(!iq->dsns_point) {
2177 /* initialize */
2178 iq->dsns_point = iq->qchase.qname;
2179 iq->dsns_point_len = iq->qchase.qname_len;
2180 }
2181 /* robustcheck for internal error: we are not underneath the dp */
2182 if(!dname_subdomain_c(iq->dsns_point, iq->dp->name)) {
2183 errinf_dname(qstate, "for DS query parent-child nameserver search the query is not under the zone", iq->dp->name);
2184 return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
2185 }
2186
2187 /* go up one (more) step, until we hit the dp, if so, end */
2188 dname_remove_label(&iq->dsns_point, &iq->dsns_point_len);
2189 if(query_dname_compare(iq->dsns_point, iq->dp->name) == 0) {
2190 /* there was no inbetween nameserver, use the old delegation
2191 * point again. And this time, because dsns_point is nonNULL
2192 * we are going to accept the (bad) result */
2193 iq->state = QUERYTARGETS_STATE;
2194 return 1;
2195 }
2196 iq->state = DSNS_FIND_STATE;
2197
2198 /* spawn NS lookup (validation not needed, this is for DS lookup) */
2199 log_nametypeclass(VERB_ALGO, "fetch nameservers",
2200 iq->dsns_point, LDNS_RR_TYPE_NS, iq->qchase.qclass);
2201 if(!generate_sub_request(iq->dsns_point, iq->dsns_point_len,
2202 LDNS_RR_TYPE_NS, iq->qchase.qclass, qstate, id, iq,
2203 INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0, 0)) {
2204 errinf_dname(qstate, "for DS query parent-child nameserver search, could not generate NS lookup for", iq->dsns_point);
2205 return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
2206 }
2207
2208 return 0;
2209 }
2210
2211 /**
2212 * Check if we wait responses for sent queries and update the iterator's
2213 * external state.
2214 */
2215 static void
check_waiting_queries(struct iter_qstate * iq,struct module_qstate * qstate,int id)2216 check_waiting_queries(struct iter_qstate* iq, struct module_qstate* qstate,
2217 int id)
2218 {
2219 if(iq->num_target_queries>0 && iq->num_current_queries>0) {
2220 verbose(VERB_ALGO, "waiting for %d targets to "
2221 "resolve or %d outstanding queries to "
2222 "respond", iq->num_target_queries,
2223 iq->num_current_queries);
2224 qstate->ext_state[id] = module_wait_reply;
2225 } else if(iq->num_target_queries>0) {
2226 verbose(VERB_ALGO, "waiting for %d targets to "
2227 "resolve", iq->num_target_queries);
2228 qstate->ext_state[id] = module_wait_subquery;
2229 } else {
2230 verbose(VERB_ALGO, "waiting for %d "
2231 "outstanding queries to respond",
2232 iq->num_current_queries);
2233 qstate->ext_state[id] = module_wait_reply;
2234 }
2235 }
2236
2237 /**
2238 * This is the request event state where the request will be sent to one of
2239 * its current query targets. This state also handles issuing target lookup
2240 * queries for missing target IP addresses. Queries typically iterate on
2241 * this state, both when they are just trying different targets for a given
2242 * delegation point, and when they change delegation points. This state
2243 * roughly corresponds to RFC 1034 algorithm steps 3 and 4.
2244 *
2245 * @param qstate: query state.
2246 * @param iq: iterator query state.
2247 * @param ie: iterator shared global environment.
2248 * @param id: module id.
2249 * @return true if the event requires more request processing immediately,
2250 * false if not. This state only returns true when it is generating
2251 * a SERVFAIL response because the query has hit a dead end.
2252 */
2253 static int
processQueryTargets(struct module_qstate * qstate,struct iter_qstate * iq,struct iter_env * ie,int id)2254 processQueryTargets(struct module_qstate* qstate, struct iter_qstate* iq,
2255 struct iter_env* ie, int id)
2256 {
2257 int tf_policy;
2258 struct delegpt_addr* target;
2259 struct outbound_entry* outq;
2260 int auth_fallback = 0;
2261 uint8_t* qout_orig = NULL;
2262 size_t qout_orig_len = 0;
2263 int sq_check_ratelimit = 1;
2264 int sq_was_ratelimited = 0;
2265
2266 /* NOTE: a request will encounter this state for each target it
2267 * needs to send a query to. That is, at least one per referral,
2268 * more if some targets timeout or return throwaway answers. */
2269
2270 log_query_info(VERB_QUERY, "processQueryTargets:", &qstate->qinfo);
2271 verbose(VERB_ALGO, "processQueryTargets: targetqueries %d, "
2272 "currentqueries %d sentcount %d", iq->num_target_queries,
2273 iq->num_current_queries, iq->sent_count);
2274
2275 /* Make sure that we haven't run away */
2276 /* FIXME: is this check even necessary? */
2277 if(iq->referral_count > MAX_REFERRAL_COUNT) {
2278 verbose(VERB_QUERY, "request has exceeded the maximum "
2279 "number of referrrals with %d", iq->referral_count);
2280 errinf(qstate, "exceeded the maximum of referrals");
2281 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2282 }
2283 if(iq->sent_count > MAX_SENT_COUNT) {
2284 verbose(VERB_QUERY, "request has exceeded the maximum "
2285 "number of sends with %d", iq->sent_count);
2286 errinf(qstate, "exceeded the maximum number of sends");
2287 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2288 }
2289
2290 /* Check if we reached MAX_TARGET_NX limit without a fallback activation. */
2291 if(iq->target_count && !*iq->nxns_dp &&
2292 iq->target_count[TARGET_COUNT_NX] > MAX_TARGET_NX) {
2293 struct delegpt_ns* ns;
2294 /* If we can wait for resolution, do so. */
2295 if(iq->num_target_queries>0 || iq->num_current_queries>0) {
2296 check_waiting_queries(iq, qstate, id);
2297 return 0;
2298 }
2299 verbose(VERB_ALGO, "request has exceeded the maximum "
2300 "number of nxdomain nameserver lookups (%d) with %d",
2301 MAX_TARGET_NX, iq->target_count[TARGET_COUNT_NX]);
2302 /* Check for dp because we require one below */
2303 if(!iq->dp) {
2304 verbose(VERB_QUERY, "Failed to get a delegation, "
2305 "giving up");
2306 errinf(qstate, "failed to get a delegation (eg. prime "
2307 "failure)");
2308 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2309 }
2310 /* We reached the limit but we already have parent side
2311 * information; stop resolution */
2312 if(iq->dp->has_parent_side_NS) {
2313 verbose(VERB_ALGO, "parent-side information is "
2314 "already present for the delegation point, no "
2315 "fallback possible");
2316 errinf(qstate, "exceeded the maximum nameserver nxdomains");
2317 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2318 }
2319 verbose(VERB_ALGO, "initiating parent-side fallback for "
2320 "nxdomain nameserver lookups");
2321 /* Mark all the current NSes as resolved to allow for parent
2322 * fallback */
2323 for(ns=iq->dp->nslist; ns; ns=ns->next) {
2324 ns->resolved = 1;
2325 }
2326 /* Note the delegation point that triggered the NXNS fallback;
2327 * no reason for shared queries to keep trying there.
2328 * This also marks the fallback activation. */
2329 *iq->nxns_dp = malloc(iq->dp->namelen);
2330 if(!*iq->nxns_dp) {
2331 verbose(VERB_ALGO, "out of memory while initiating "
2332 "fallback");
2333 errinf(qstate, "exceeded the maximum nameserver "
2334 "nxdomains (malloc)");
2335 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2336 }
2337 memcpy(*iq->nxns_dp, iq->dp->name, iq->dp->namelen);
2338 } else if(iq->target_count && *iq->nxns_dp) {
2339 /* Handle the NXNS fallback case. */
2340 /* If we can wait for resolution, do so. */
2341 if(iq->num_target_queries>0 || iq->num_current_queries>0) {
2342 check_waiting_queries(iq, qstate, id);
2343 return 0;
2344 }
2345 /* Check for dp because we require one below */
2346 if(!iq->dp) {
2347 verbose(VERB_QUERY, "Failed to get a delegation, "
2348 "giving up");
2349 errinf(qstate, "failed to get a delegation (eg. prime "
2350 "failure)");
2351 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2352 }
2353
2354 if(iq->target_count[TARGET_COUNT_NX] > MAX_TARGET_NX_FALLBACK) {
2355 verbose(VERB_ALGO, "request has exceeded the maximum "
2356 "number of fallback nxdomain nameserver "
2357 "lookups (%d) with %d", MAX_TARGET_NX_FALLBACK,
2358 iq->target_count[TARGET_COUNT_NX]);
2359 errinf(qstate, "exceeded the maximum nameserver nxdomains");
2360 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2361 }
2362
2363 if(!iq->dp->has_parent_side_NS) {
2364 struct delegpt_ns* ns;
2365 if(!dname_canonical_compare(*iq->nxns_dp, iq->dp->name)) {
2366 verbose(VERB_ALGO, "this delegation point "
2367 "initiated the fallback, marking the "
2368 "nslist as resolved");
2369 for(ns=iq->dp->nslist; ns; ns=ns->next) {
2370 ns->resolved = 1;
2371 }
2372 }
2373 }
2374 }
2375
2376 /* Make sure we have a delegation point, otherwise priming failed
2377 * or another failure occurred */
2378 if(!iq->dp) {
2379 verbose(VERB_QUERY, "Failed to get a delegation, giving up");
2380 errinf(qstate, "failed to get a delegation (eg. prime failure)");
2381 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2382 }
2383 if(!ie->supports_ipv6)
2384 delegpt_no_ipv6(iq->dp);
2385 if(!ie->supports_ipv4)
2386 delegpt_no_ipv4(iq->dp);
2387 delegpt_log(VERB_ALGO, iq->dp);
2388
2389 if(iq->num_current_queries>0) {
2390 /* already busy answering a query, this restart is because
2391 * more delegpt addrs became available, wait for existing
2392 * query. */
2393 verbose(VERB_ALGO, "woke up, but wait for outstanding query");
2394 qstate->ext_state[id] = module_wait_reply;
2395 return 0;
2396 }
2397
2398 if(iq->minimisation_state == INIT_MINIMISE_STATE
2399 && !(iq->chase_flags & BIT_RD)) {
2400 /* (Re)set qinfo_out to (new) delegation point, except when
2401 * qinfo_out is already a subdomain of dp. This happens when
2402 * increasing by more than one label at once (QNAMEs with more
2403 * than MAX_MINIMISE_COUNT labels). */
2404 if(!(iq->qinfo_out.qname_len
2405 && dname_subdomain_c(iq->qchase.qname,
2406 iq->qinfo_out.qname)
2407 && dname_subdomain_c(iq->qinfo_out.qname,
2408 iq->dp->name))) {
2409 iq->qinfo_out.qname = iq->dp->name;
2410 iq->qinfo_out.qname_len = iq->dp->namelen;
2411 iq->qinfo_out.qtype = LDNS_RR_TYPE_A;
2412 iq->qinfo_out.qclass = iq->qchase.qclass;
2413 iq->qinfo_out.local_alias = NULL;
2414 iq->minimise_count = 0;
2415 }
2416
2417 iq->minimisation_state = MINIMISE_STATE;
2418 }
2419 if(iq->minimisation_state == MINIMISE_STATE) {
2420 int qchaselabs = dname_count_labels(iq->qchase.qname);
2421 int labdiff = qchaselabs -
2422 dname_count_labels(iq->qinfo_out.qname);
2423
2424 qout_orig = iq->qinfo_out.qname;
2425 qout_orig_len = iq->qinfo_out.qname_len;
2426 iq->qinfo_out.qname = iq->qchase.qname;
2427 iq->qinfo_out.qname_len = iq->qchase.qname_len;
2428 iq->minimise_count++;
2429 iq->timeout_count = 0;
2430
2431 iter_dec_attempts(iq->dp, 1, ie->outbound_msg_retry);
2432
2433 /* Limit number of iterations for QNAMEs with more
2434 * than MAX_MINIMISE_COUNT labels. Send first MINIMISE_ONE_LAB
2435 * labels of QNAME always individually.
2436 */
2437 if(qchaselabs > MAX_MINIMISE_COUNT && labdiff > 1 &&
2438 iq->minimise_count > MINIMISE_ONE_LAB) {
2439 if(iq->minimise_count < MAX_MINIMISE_COUNT) {
2440 int multilabs = qchaselabs - 1 -
2441 MINIMISE_ONE_LAB;
2442 int extralabs = multilabs /
2443 MINIMISE_MULTIPLE_LABS;
2444
2445 if (MAX_MINIMISE_COUNT - iq->minimise_count >=
2446 multilabs % MINIMISE_MULTIPLE_LABS)
2447 /* Default behaviour is to add 1 label
2448 * every iteration. Therefore, decrement
2449 * the extralabs by 1 */
2450 extralabs--;
2451 if (extralabs < labdiff)
2452 labdiff -= extralabs;
2453 else
2454 labdiff = 1;
2455 }
2456 /* Last minimised iteration, send all labels with
2457 * QTYPE=NS */
2458 else
2459 labdiff = 1;
2460 }
2461
2462 if(labdiff > 1) {
2463 verbose(VERB_QUERY, "removing %d labels", labdiff-1);
2464 dname_remove_labels(&iq->qinfo_out.qname,
2465 &iq->qinfo_out.qname_len,
2466 labdiff-1);
2467 }
2468 if(labdiff < 1 || (labdiff < 2
2469 && (iq->qchase.qtype == LDNS_RR_TYPE_DS
2470 || iq->qchase.qtype == LDNS_RR_TYPE_A)))
2471 /* Stop minimising this query, resolve "as usual" */
2472 iq->minimisation_state = DONOT_MINIMISE_STATE;
2473 else if(!qstate->no_cache_lookup) {
2474 struct dns_msg* msg = dns_cache_lookup(qstate->env,
2475 iq->qinfo_out.qname, iq->qinfo_out.qname_len,
2476 iq->qinfo_out.qtype, iq->qinfo_out.qclass,
2477 qstate->query_flags, qstate->region,
2478 qstate->env->scratch, 0, iq->dp->name,
2479 iq->dp->namelen);
2480 if(msg && FLAGS_GET_RCODE(msg->rep->flags) ==
2481 LDNS_RCODE_NOERROR)
2482 /* no need to send query if it is already
2483 * cached as NOERROR */
2484 return 1;
2485 if(msg && FLAGS_GET_RCODE(msg->rep->flags) ==
2486 LDNS_RCODE_NXDOMAIN &&
2487 qstate->env->need_to_validate &&
2488 qstate->env->cfg->harden_below_nxdomain) {
2489 if(msg->rep->security == sec_status_secure) {
2490 iq->response = msg;
2491 return final_state(iq);
2492 }
2493 if(msg->rep->security == sec_status_unchecked) {
2494 struct module_qstate* subq = NULL;
2495 if(!generate_sub_request(
2496 iq->qinfo_out.qname,
2497 iq->qinfo_out.qname_len,
2498 iq->qinfo_out.qtype,
2499 iq->qinfo_out.qclass,
2500 qstate, id, iq,
2501 INIT_REQUEST_STATE,
2502 FINISHED_STATE, &subq, 1, 1))
2503 verbose(VERB_ALGO,
2504 "could not validate NXDOMAIN "
2505 "response");
2506 }
2507 }
2508 if(msg && FLAGS_GET_RCODE(msg->rep->flags) ==
2509 LDNS_RCODE_NXDOMAIN) {
2510 /* return and add a label in the next
2511 * minimisation iteration.
2512 */
2513 return 1;
2514 }
2515 }
2516 }
2517 if(iq->minimisation_state == SKIP_MINIMISE_STATE) {
2518 if(iq->timeout_count < MAX_MINIMISE_TIMEOUT_COUNT)
2519 /* Do not increment qname, continue incrementing next
2520 * iteration */
2521 iq->minimisation_state = MINIMISE_STATE;
2522 else if(!qstate->env->cfg->qname_minimisation_strict)
2523 /* Too many time-outs detected for this QNAME and QTYPE.
2524 * We give up, disable QNAME minimisation. */
2525 iq->minimisation_state = DONOT_MINIMISE_STATE;
2526 }
2527 if(iq->minimisation_state == DONOT_MINIMISE_STATE)
2528 iq->qinfo_out = iq->qchase;
2529
2530 /* now find an answer to this query */
2531 /* see if authority zones have an answer */
2532 /* now we know the dp, we can check the auth zone for locally hosted
2533 * contents */
2534 if(!iq->auth_zone_avoid && qstate->blacklist) {
2535 if(auth_zones_can_fallback(qstate->env->auth_zones,
2536 iq->dp->name, iq->dp->namelen, iq->qinfo_out.qclass)) {
2537 /* if cache is blacklisted and this zone allows us
2538 * to fallback to the internet, then do so, and
2539 * fetch results from the internet servers */
2540 iq->auth_zone_avoid = 1;
2541 }
2542 }
2543 if(iq->auth_zone_avoid) {
2544 iq->auth_zone_avoid = 0;
2545 auth_fallback = 1;
2546 } else if(auth_zones_lookup(qstate->env->auth_zones, &iq->qinfo_out,
2547 qstate->region, &iq->response, &auth_fallback, iq->dp->name,
2548 iq->dp->namelen)) {
2549 /* use this as a response to be processed by the iterator */
2550 if(verbosity >= VERB_ALGO) {
2551 log_dns_msg("msg from auth zone",
2552 &iq->response->qinfo, iq->response->rep);
2553 }
2554 if((iq->chase_flags&BIT_RD) && !(iq->response->rep->flags&BIT_AA)) {
2555 verbose(VERB_ALGO, "forwarder, ignoring referral from auth zone");
2556 } else {
2557 lock_rw_wrlock(&qstate->env->auth_zones->lock);
2558 qstate->env->auth_zones->num_query_up++;
2559 lock_rw_unlock(&qstate->env->auth_zones->lock);
2560 iq->num_current_queries++;
2561 iq->chase_to_rd = 0;
2562 iq->dnssec_lame_query = 0;
2563 iq->auth_zone_response = 1;
2564 return next_state(iq, QUERY_RESP_STATE);
2565 }
2566 }
2567 iq->auth_zone_response = 0;
2568 if(auth_fallback == 0) {
2569 /* like we got servfail from the auth zone lookup, and
2570 * no internet fallback */
2571 verbose(VERB_ALGO, "auth zone lookup failed, no fallback,"
2572 " servfail");
2573 errinf(qstate, "auth zone lookup failed, fallback is off");
2574 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2575 }
2576 if(iq->dp->auth_dp) {
2577 /* we wanted to fallback, but had no delegpt, only the
2578 * auth zone generated delegpt, create an actual one */
2579 iq->auth_zone_avoid = 1;
2580 return next_state(iq, INIT_REQUEST_STATE);
2581 }
2582 /* but mostly, fallback==1 (like, when no such auth zone exists)
2583 * and we continue with lookups */
2584
2585 tf_policy = 0;
2586 /* < not <=, because although the array is large enough for <=, the
2587 * generated query will immediately be discarded due to depth and
2588 * that servfail is cached, which is not good as opportunism goes. */
2589 if(iq->depth < ie->max_dependency_depth
2590 && iq->num_target_queries == 0
2591 && (!iq->target_count || iq->target_count[TARGET_COUNT_NX]==0)
2592 && iq->sent_count < TARGET_FETCH_STOP
2593 /* if the mesh query list is full, then do not waste cpu
2594 * and sockets to fetch promiscuous targets. They can be
2595 * looked up when needed. */
2596 && !mesh_jostle_exceeded(qstate->env->mesh)
2597 ) {
2598 tf_policy = ie->target_fetch_policy[iq->depth];
2599 }
2600
2601 /* if in 0x20 fallback get as many targets as possible */
2602 if(iq->caps_fallback) {
2603 int extra = 0;
2604 size_t naddr, nres, navail;
2605 if(!query_for_targets(qstate, iq, ie, id, -1, &extra)) {
2606 errinf(qstate, "could not fetch nameservers for 0x20 fallback");
2607 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2608 }
2609 iq->num_target_queries += extra;
2610 target_count_increase(iq, extra);
2611 if(iq->num_target_queries > 0) {
2612 /* wait to get all targets, we want to try em */
2613 verbose(VERB_ALGO, "wait for all targets for fallback");
2614 qstate->ext_state[id] = module_wait_reply;
2615 /* undo qname minimise step because we'll get back here
2616 * to do it again */
2617 if(qout_orig && iq->minimise_count > 0) {
2618 iq->minimise_count--;
2619 iq->qinfo_out.qname = qout_orig;
2620 iq->qinfo_out.qname_len = qout_orig_len;
2621 }
2622 return 0;
2623 }
2624 /* did we do enough fallback queries already? */
2625 delegpt_count_addr(iq->dp, &naddr, &nres, &navail);
2626 /* the current caps_server is the number of fallbacks sent.
2627 * the original query is one that matched too, so we have
2628 * caps_server+1 number of matching queries now */
2629 if(iq->caps_server+1 >= naddr*3 ||
2630 iq->caps_server*2+2 >= MAX_SENT_COUNT) {
2631 /* *2 on sentcount check because ipv6 may fail */
2632 /* we're done, process the response */
2633 verbose(VERB_ALGO, "0x20 fallback had %d responses "
2634 "match for %d wanted, done.",
2635 (int)iq->caps_server+1, (int)naddr*3);
2636 iq->response = iq->caps_response;
2637 iq->caps_fallback = 0;
2638 iter_dec_attempts(iq->dp, 3, ie->outbound_msg_retry); /* space for fallback */
2639 iq->num_current_queries++; /* RespState decrements it*/
2640 iq->referral_count++; /* make sure we don't loop */
2641 iq->sent_count = 0;
2642 iq->dp_target_count = 0;
2643 iq->state = QUERY_RESP_STATE;
2644 return 1;
2645 }
2646 verbose(VERB_ALGO, "0x20 fallback number %d",
2647 (int)iq->caps_server);
2648
2649 /* if there is a policy to fetch missing targets
2650 * opportunistically, do it. we rely on the fact that once a
2651 * query (or queries) for a missing name have been issued,
2652 * they will not show up again. */
2653 } else if(tf_policy != 0) {
2654 int extra = 0;
2655 verbose(VERB_ALGO, "attempt to get extra %d targets",
2656 tf_policy);
2657 (void)query_for_targets(qstate, iq, ie, id, tf_policy, &extra);
2658 /* errors ignored, these targets are not strictly necessary for
2659 * this result, we do not have to reply with SERVFAIL */
2660 iq->num_target_queries += extra;
2661 target_count_increase(iq, extra);
2662 }
2663
2664 /* Add the current set of unused targets to our queue. */
2665 delegpt_add_unused_targets(iq->dp);
2666
2667 if(qstate->env->auth_zones) {
2668 /* apply rpz triggers at query time */
2669 struct dns_msg* forged_response = rpz_callback_from_iterator_module(qstate, iq);
2670 if(forged_response != NULL) {
2671 qstate->ext_state[id] = module_finished;
2672 qstate->return_rcode = LDNS_RCODE_NOERROR;
2673 qstate->return_msg = forged_response;
2674 iq->response = forged_response;
2675 next_state(iq, FINISHED_STATE);
2676 if(!iter_prepend(iq, qstate->return_msg, qstate->region)) {
2677 log_err("rpz: prepend rrsets: out of memory");
2678 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2679 }
2680 return 0;
2681 }
2682 }
2683
2684 /* Select the next usable target, filtering out unsuitable targets. */
2685 target = iter_server_selection(ie, qstate->env, iq->dp,
2686 iq->dp->name, iq->dp->namelen, iq->qchase.qtype,
2687 &iq->dnssec_lame_query, &iq->chase_to_rd,
2688 iq->num_target_queries, qstate->blacklist,
2689 qstate->prefetch_leeway);
2690
2691 /* If no usable target was selected... */
2692 if(!target) {
2693 /* Here we distinguish between three states: generate a new
2694 * target query, just wait, or quit (with a SERVFAIL).
2695 * We have the following information: number of active
2696 * target queries, number of active current queries,
2697 * the presence of missing targets at this delegation
2698 * point, and the given query target policy. */
2699
2700 /* Check for the wait condition. If this is true, then
2701 * an action must be taken. */
2702 if(iq->num_target_queries==0 && iq->num_current_queries==0) {
2703 /* If there is nothing to wait for, then we need
2704 * to distinguish between generating (a) new target
2705 * query, or failing. */
2706 if(delegpt_count_missing_targets(iq->dp, NULL) > 0) {
2707 int qs = 0;
2708 verbose(VERB_ALGO, "querying for next "
2709 "missing target");
2710 if(!query_for_targets(qstate, iq, ie, id,
2711 1, &qs)) {
2712 errinf(qstate, "could not fetch nameserver");
2713 errinf_dname(qstate, "at zone", iq->dp->name);
2714 return error_response(qstate, id,
2715 LDNS_RCODE_SERVFAIL);
2716 }
2717 if(qs == 0 &&
2718 delegpt_count_missing_targets(iq->dp, NULL) == 0){
2719 /* it looked like there were missing
2720 * targets, but they did not turn up.
2721 * Try the bad choices again (if any),
2722 * when we get back here missing==0,
2723 * so this is not a loop. */
2724 return 1;
2725 }
2726 iq->num_target_queries += qs;
2727 target_count_increase(iq, qs);
2728 }
2729 /* Since a target query might have been made, we
2730 * need to check again. */
2731 if(iq->num_target_queries == 0) {
2732 /* if in capsforid fallback, instead of last
2733 * resort, we agree with the current reply
2734 * we have (if any) (our count of addrs bad)*/
2735 if(iq->caps_fallback && iq->caps_reply) {
2736 /* we're done, process the response */
2737 verbose(VERB_ALGO, "0x20 fallback had %d responses, "
2738 "but no more servers except "
2739 "last resort, done.",
2740 (int)iq->caps_server+1);
2741 iq->response = iq->caps_response;
2742 iq->caps_fallback = 0;
2743 iter_dec_attempts(iq->dp, 3, ie->outbound_msg_retry); /* space for fallback */
2744 iq->num_current_queries++; /* RespState decrements it*/
2745 iq->referral_count++; /* make sure we don't loop */
2746 iq->sent_count = 0;
2747 iq->dp_target_count = 0;
2748 iq->state = QUERY_RESP_STATE;
2749 return 1;
2750 }
2751 return processLastResort(qstate, iq, ie, id);
2752 }
2753 }
2754
2755 /* otherwise, we have no current targets, so submerge
2756 * until one of the target or direct queries return. */
2757 verbose(VERB_ALGO, "no current targets");
2758 check_waiting_queries(iq, qstate, id);
2759 /* undo qname minimise step because we'll get back here
2760 * to do it again */
2761 if(qout_orig && iq->minimise_count > 0) {
2762 iq->minimise_count--;
2763 iq->qinfo_out.qname = qout_orig;
2764 iq->qinfo_out.qname_len = qout_orig_len;
2765 }
2766 return 0;
2767 }
2768
2769 /* Do not check ratelimit for forwarding queries or if we already got a
2770 * pass. */
2771 sq_check_ratelimit = (!(iq->chase_flags & BIT_RD) && !iq->ratelimit_ok);
2772 /* We have a valid target. */
2773 if(verbosity >= VERB_QUERY) {
2774 log_query_info(VERB_QUERY, "sending query:", &iq->qinfo_out);
2775 log_name_addr(VERB_QUERY, "sending to target:", iq->dp->name,
2776 &target->addr, target->addrlen);
2777 verbose(VERB_ALGO, "dnssec status: %s%s",
2778 iq->dnssec_expected?"expected": "not expected",
2779 iq->dnssec_lame_query?" but lame_query anyway": "");
2780 }
2781 fptr_ok(fptr_whitelist_modenv_send_query(qstate->env->send_query));
2782 outq = (*qstate->env->send_query)(&iq->qinfo_out,
2783 iq->chase_flags | (iq->chase_to_rd?BIT_RD:0),
2784 /* unset CD if to forwarder(RD set) and not dnssec retry
2785 * (blacklist nonempty) and no trust-anchors are configured
2786 * above the qname or on the first attempt when dnssec is on */
2787 EDNS_DO| ((iq->chase_to_rd||(iq->chase_flags&BIT_RD)!=0)&&
2788 !qstate->blacklist&&(!iter_qname_indicates_dnssec(qstate->env,
2789 &iq->qinfo_out)||target->attempts==1)?0:BIT_CD),
2790 iq->dnssec_expected, iq->caps_fallback || is_caps_whitelisted(
2791 ie, iq), sq_check_ratelimit, &target->addr, target->addrlen,
2792 iq->dp->name, iq->dp->namelen,
2793 (iq->dp->tcp_upstream || qstate->env->cfg->tcp_upstream),
2794 (iq->dp->ssl_upstream || qstate->env->cfg->ssl_upstream),
2795 target->tls_auth_name, qstate, &sq_was_ratelimited);
2796 if(!outq) {
2797 if(sq_was_ratelimited) {
2798 lock_basic_lock(&ie->queries_ratelimit_lock);
2799 ie->num_queries_ratelimited++;
2800 lock_basic_unlock(&ie->queries_ratelimit_lock);
2801 verbose(VERB_ALGO, "query exceeded ratelimits");
2802 qstate->was_ratelimited = 1;
2803 errinf_dname(qstate, "exceeded ratelimit for zone",
2804 iq->dp->name);
2805 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2806 }
2807 log_addr(VERB_QUERY, "error sending query to auth server",
2808 &target->addr, target->addrlen);
2809 if(qstate->env->cfg->qname_minimisation)
2810 iq->minimisation_state = SKIP_MINIMISE_STATE;
2811 return next_state(iq, QUERYTARGETS_STATE);
2812 }
2813 outbound_list_insert(&iq->outlist, outq);
2814 iq->num_current_queries++;
2815 iq->sent_count++;
2816 qstate->ext_state[id] = module_wait_reply;
2817
2818 return 0;
2819 }
2820
2821 /** find NS rrset in given list */
2822 static struct ub_packed_rrset_key*
find_NS(struct reply_info * rep,size_t from,size_t to)2823 find_NS(struct reply_info* rep, size_t from, size_t to)
2824 {
2825 size_t i;
2826 for(i=from; i<to; i++) {
2827 if(ntohs(rep->rrsets[i]->rk.type) == LDNS_RR_TYPE_NS)
2828 return rep->rrsets[i];
2829 }
2830 return NULL;
2831 }
2832
2833
2834 /**
2835 * Process the query response. All queries end up at this state first. This
2836 * process generally consists of analyzing the response and routing the
2837 * event to the next state (either bouncing it back to a request state, or
2838 * terminating the processing for this event).
2839 *
2840 * @param qstate: query state.
2841 * @param iq: iterator query state.
2842 * @param ie: iterator shared global environment.
2843 * @param id: module id.
2844 * @return true if the event requires more immediate processing, false if
2845 * not. This is generally only true when forwarding the request to
2846 * the final state (i.e., on answer).
2847 */
2848 static int
processQueryResponse(struct module_qstate * qstate,struct iter_qstate * iq,struct iter_env * ie,int id)2849 processQueryResponse(struct module_qstate* qstate, struct iter_qstate* iq,
2850 struct iter_env* ie, int id)
2851 {
2852 int dnsseclame = 0;
2853 enum response_type type;
2854
2855 iq->num_current_queries--;
2856
2857 if(!inplace_cb_query_response_call(qstate->env, qstate, iq->response))
2858 log_err("unable to call query_response callback");
2859
2860 if(iq->response == NULL) {
2861 /* Don't increment qname when QNAME minimisation is enabled */
2862 if(qstate->env->cfg->qname_minimisation) {
2863 iq->minimisation_state = SKIP_MINIMISE_STATE;
2864 }
2865 iq->timeout_count++;
2866 iq->chase_to_rd = 0;
2867 iq->dnssec_lame_query = 0;
2868 verbose(VERB_ALGO, "query response was timeout");
2869 return next_state(iq, QUERYTARGETS_STATE);
2870 }
2871 iq->timeout_count = 0;
2872 type = response_type_from_server(
2873 (int)((iq->chase_flags&BIT_RD) || iq->chase_to_rd),
2874 iq->response, &iq->qinfo_out, iq->dp);
2875 iq->chase_to_rd = 0;
2876 if(type == RESPONSE_TYPE_REFERRAL && (iq->chase_flags&BIT_RD) &&
2877 !iq->auth_zone_response) {
2878 /* When forwarding (RD bit is set), we handle referrals
2879 * differently. No queries should be sent elsewhere */
2880 type = RESPONSE_TYPE_ANSWER;
2881 }
2882 if(!qstate->env->cfg->disable_dnssec_lame_check && iq->dnssec_expected
2883 && !iq->dnssec_lame_query &&
2884 !(iq->chase_flags&BIT_RD)
2885 && iq->sent_count < DNSSEC_LAME_DETECT_COUNT
2886 && type != RESPONSE_TYPE_LAME
2887 && type != RESPONSE_TYPE_REC_LAME
2888 && type != RESPONSE_TYPE_THROWAWAY
2889 && type != RESPONSE_TYPE_UNTYPED) {
2890 /* a possible answer, see if it is missing DNSSEC */
2891 /* but not when forwarding, so we dont mark fwder lame */
2892 if(!iter_msg_has_dnssec(iq->response)) {
2893 /* Mark this address as dnsseclame in this dp,
2894 * because that will make serverselection disprefer
2895 * it, but also, once it is the only final option,
2896 * use dnssec-lame-bypass if it needs to query there.*/
2897 if(qstate->reply) {
2898 struct delegpt_addr* a = delegpt_find_addr(
2899 iq->dp, &qstate->reply->addr,
2900 qstate->reply->addrlen);
2901 if(a) a->dnsseclame = 1;
2902 }
2903 /* test the answer is from the zone we expected,
2904 * otherwise, (due to parent,child on same server), we
2905 * might mark the server,zone lame inappropriately */
2906 if(!iter_msg_from_zone(iq->response, iq->dp, type,
2907 iq->qchase.qclass))
2908 qstate->reply = NULL;
2909 type = RESPONSE_TYPE_LAME;
2910 dnsseclame = 1;
2911 }
2912 } else iq->dnssec_lame_query = 0;
2913 /* see if referral brings us close to the target */
2914 if(type == RESPONSE_TYPE_REFERRAL) {
2915 struct ub_packed_rrset_key* ns = find_NS(
2916 iq->response->rep, iq->response->rep->an_numrrsets,
2917 iq->response->rep->an_numrrsets
2918 + iq->response->rep->ns_numrrsets);
2919 if(!ns) ns = find_NS(iq->response->rep, 0,
2920 iq->response->rep->an_numrrsets);
2921 if(!ns || !dname_strict_subdomain_c(ns->rk.dname, iq->dp->name)
2922 || !dname_subdomain_c(iq->qchase.qname, ns->rk.dname)){
2923 verbose(VERB_ALGO, "bad referral, throwaway");
2924 type = RESPONSE_TYPE_THROWAWAY;
2925 } else
2926 iter_scrub_ds(iq->response, ns, iq->dp->name);
2927 } else iter_scrub_ds(iq->response, NULL, NULL);
2928 if(type == RESPONSE_TYPE_THROWAWAY &&
2929 FLAGS_GET_RCODE(iq->response->rep->flags) == LDNS_RCODE_YXDOMAIN) {
2930 /* YXDOMAIN is a permanent error, no need to retry */
2931 type = RESPONSE_TYPE_ANSWER;
2932 }
2933 if(type == RESPONSE_TYPE_CNAME && iq->response->rep->an_numrrsets >= 1
2934 && ntohs(iq->response->rep->rrsets[0]->rk.type) == LDNS_RR_TYPE_DNAME) {
2935 uint8_t* sname = NULL;
2936 size_t snamelen = 0;
2937 get_cname_target(iq->response->rep->rrsets[0], &sname,
2938 &snamelen);
2939 if(snamelen && dname_subdomain_c(sname, iq->response->rep->rrsets[0]->rk.dname)) {
2940 /* DNAME to a subdomain loop; do not recurse */
2941 type = RESPONSE_TYPE_ANSWER;
2942 }
2943 } else if(type == RESPONSE_TYPE_CNAME &&
2944 iq->qchase.qtype == LDNS_RR_TYPE_CNAME &&
2945 iq->minimisation_state == MINIMISE_STATE &&
2946 query_dname_compare(iq->qchase.qname, iq->qinfo_out.qname) == 0) {
2947 /* The minimised query for full QTYPE and hidden QTYPE can be
2948 * classified as CNAME response type, even when the original
2949 * QTYPE=CNAME. This should be treated as answer response type.
2950 */
2951 type = RESPONSE_TYPE_ANSWER;
2952 }
2953
2954 /* handle each of the type cases */
2955 if(type == RESPONSE_TYPE_ANSWER) {
2956 /* ANSWER type responses terminate the query algorithm,
2957 * so they sent on their */
2958 if(verbosity >= VERB_DETAIL) {
2959 verbose(VERB_DETAIL, "query response was %s",
2960 FLAGS_GET_RCODE(iq->response->rep->flags)
2961 ==LDNS_RCODE_NXDOMAIN?"NXDOMAIN ANSWER":
2962 (iq->response->rep->an_numrrsets?"ANSWER":
2963 "nodata ANSWER"));
2964 }
2965 /* if qtype is DS, check we have the right level of answer,
2966 * like grandchild answer but we need the middle, reject it */
2967 if(iq->qchase.qtype == LDNS_RR_TYPE_DS && !iq->dsns_point
2968 && !(iq->chase_flags&BIT_RD)
2969 && iter_ds_toolow(iq->response, iq->dp)
2970 && iter_dp_cangodown(&iq->qchase, iq->dp)) {
2971 /* close down outstanding requests to be discarded */
2972 outbound_list_clear(&iq->outlist);
2973 iq->num_current_queries = 0;
2974 fptr_ok(fptr_whitelist_modenv_detach_subs(
2975 qstate->env->detach_subs));
2976 (*qstate->env->detach_subs)(qstate);
2977 iq->num_target_queries = 0;
2978 return processDSNSFind(qstate, iq, id);
2979 }
2980 if(!qstate->no_cache_store)
2981 iter_dns_store(qstate->env, &iq->response->qinfo,
2982 iq->response->rep,
2983 iq->qchase.qtype != iq->response->qinfo.qtype,
2984 qstate->prefetch_leeway,
2985 iq->dp&&iq->dp->has_parent_side_NS,
2986 qstate->region, qstate->query_flags,
2987 qstate->qstarttime);
2988 /* close down outstanding requests to be discarded */
2989 outbound_list_clear(&iq->outlist);
2990 iq->num_current_queries = 0;
2991 fptr_ok(fptr_whitelist_modenv_detach_subs(
2992 qstate->env->detach_subs));
2993 (*qstate->env->detach_subs)(qstate);
2994 iq->num_target_queries = 0;
2995 if(qstate->reply)
2996 sock_list_insert(&qstate->reply_origin,
2997 &qstate->reply->addr, qstate->reply->addrlen,
2998 qstate->region);
2999 if(iq->minimisation_state != DONOT_MINIMISE_STATE
3000 && !(iq->chase_flags & BIT_RD)) {
3001 if(FLAGS_GET_RCODE(iq->response->rep->flags) !=
3002 LDNS_RCODE_NOERROR) {
3003 if(qstate->env->cfg->qname_minimisation_strict) {
3004 if(FLAGS_GET_RCODE(iq->response->rep->flags) ==
3005 LDNS_RCODE_NXDOMAIN) {
3006 iter_scrub_nxdomain(iq->response);
3007 return final_state(iq);
3008 }
3009 return error_response(qstate, id,
3010 LDNS_RCODE_SERVFAIL);
3011 }
3012 /* Best effort qname-minimisation.
3013 * Stop minimising and send full query when
3014 * RCODE is not NOERROR. */
3015 iq->minimisation_state = DONOT_MINIMISE_STATE;
3016 }
3017 if(FLAGS_GET_RCODE(iq->response->rep->flags) ==
3018 LDNS_RCODE_NXDOMAIN) {
3019 /* Stop resolving when NXDOMAIN is DNSSEC
3020 * signed. Based on assumption that nameservers
3021 * serving signed zones do not return NXDOMAIN
3022 * for empty-non-terminals. */
3023 if(iq->dnssec_expected)
3024 return final_state(iq);
3025 /* Make subrequest to validate intermediate
3026 * NXDOMAIN if harden-below-nxdomain is
3027 * enabled. */
3028 if(qstate->env->cfg->harden_below_nxdomain &&
3029 qstate->env->need_to_validate) {
3030 struct module_qstate* subq = NULL;
3031 log_query_info(VERB_QUERY,
3032 "schedule NXDOMAIN validation:",
3033 &iq->response->qinfo);
3034 if(!generate_sub_request(
3035 iq->response->qinfo.qname,
3036 iq->response->qinfo.qname_len,
3037 iq->response->qinfo.qtype,
3038 iq->response->qinfo.qclass,
3039 qstate, id, iq,
3040 INIT_REQUEST_STATE,
3041 FINISHED_STATE, &subq, 1, 1))
3042 verbose(VERB_ALGO,
3043 "could not validate NXDOMAIN "
3044 "response");
3045 }
3046 }
3047 return next_state(iq, QUERYTARGETS_STATE);
3048 }
3049 return final_state(iq);
3050 } else if(type == RESPONSE_TYPE_REFERRAL) {
3051 /* REFERRAL type responses get a reset of the
3052 * delegation point, and back to the QUERYTARGETS_STATE. */
3053 verbose(VERB_DETAIL, "query response was REFERRAL");
3054
3055 /* if hardened, only store referral if we asked for it */
3056 if(!qstate->no_cache_store &&
3057 (!qstate->env->cfg->harden_referral_path ||
3058 ( qstate->qinfo.qtype == LDNS_RR_TYPE_NS
3059 && (qstate->query_flags&BIT_RD)
3060 && !(qstate->query_flags&BIT_CD)
3061 /* we know that all other NS rrsets are scrubbed
3062 * away, thus on referral only one is left.
3063 * see if that equals the query name... */
3064 && ( /* auth section, but sometimes in answer section*/
3065 reply_find_rrset_section_ns(iq->response->rep,
3066 iq->qchase.qname, iq->qchase.qname_len,
3067 LDNS_RR_TYPE_NS, iq->qchase.qclass)
3068 || reply_find_rrset_section_an(iq->response->rep,
3069 iq->qchase.qname, iq->qchase.qname_len,
3070 LDNS_RR_TYPE_NS, iq->qchase.qclass)
3071 )
3072 ))) {
3073 /* Store the referral under the current query */
3074 /* no prefetch-leeway, since its not the answer */
3075 iter_dns_store(qstate->env, &iq->response->qinfo,
3076 iq->response->rep, 1, 0, 0, NULL, 0,
3077 qstate->qstarttime);
3078 if(iq->store_parent_NS)
3079 iter_store_parentside_NS(qstate->env,
3080 iq->response->rep);
3081 if(qstate->env->neg_cache)
3082 val_neg_addreferral(qstate->env->neg_cache,
3083 iq->response->rep, iq->dp->name);
3084 }
3085 /* store parent-side-in-zone-glue, if directly queried for */
3086 if(!qstate->no_cache_store && iq->query_for_pside_glue
3087 && !iq->pside_glue) {
3088 iq->pside_glue = reply_find_rrset(iq->response->rep,
3089 iq->qchase.qname, iq->qchase.qname_len,
3090 iq->qchase.qtype, iq->qchase.qclass);
3091 if(iq->pside_glue) {
3092 log_rrset_key(VERB_ALGO, "found parent-side "
3093 "glue", iq->pside_glue);
3094 iter_store_parentside_rrset(qstate->env,
3095 iq->pside_glue);
3096 }
3097 }
3098
3099 /* Reset the event state, setting the current delegation
3100 * point to the referral. */
3101 iq->deleg_msg = iq->response;
3102 iq->dp = delegpt_from_message(iq->response, qstate->region);
3103 if (qstate->env->cfg->qname_minimisation)
3104 iq->minimisation_state = INIT_MINIMISE_STATE;
3105 if(!iq->dp) {
3106 errinf(qstate, "malloc failure, for delegation point");
3107 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3108 }
3109 if(!cache_fill_missing(qstate->env, iq->qchase.qclass,
3110 qstate->region, iq->dp)) {
3111 errinf(qstate, "malloc failure, copy extra info into delegation point");
3112 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3113 }
3114 if(iq->store_parent_NS && query_dname_compare(iq->dp->name,
3115 iq->store_parent_NS->name) == 0)
3116 iter_merge_retry_counts(iq->dp, iq->store_parent_NS,
3117 ie->outbound_msg_retry);
3118 delegpt_log(VERB_ALGO, iq->dp);
3119 /* Count this as a referral. */
3120 iq->referral_count++;
3121 iq->sent_count = 0;
3122 iq->dp_target_count = 0;
3123 /* see if the next dp is a trust anchor, or a DS was sent
3124 * along, indicating dnssec is expected for next zone */
3125 iq->dnssec_expected = iter_indicates_dnssec(qstate->env,
3126 iq->dp, iq->response, iq->qchase.qclass);
3127 /* if dnssec, validating then also fetch the key for the DS */
3128 if(iq->dnssec_expected && qstate->env->cfg->prefetch_key &&
3129 !(qstate->query_flags&BIT_CD))
3130 generate_dnskey_prefetch(qstate, iq, id);
3131
3132 /* spawn off NS and addr to auth servers for the NS we just
3133 * got in the referral. This gets authoritative answer
3134 * (answer section trust level) rrset.
3135 * right after, we detach the subs, answer goes to cache. */
3136 if(qstate->env->cfg->harden_referral_path)
3137 generate_ns_check(qstate, iq, id);
3138
3139 /* stop current outstanding queries.
3140 * FIXME: should the outstanding queries be waited for and
3141 * handled? Say by a subquery that inherits the outbound_entry.
3142 */
3143 outbound_list_clear(&iq->outlist);
3144 iq->num_current_queries = 0;
3145 fptr_ok(fptr_whitelist_modenv_detach_subs(
3146 qstate->env->detach_subs));
3147 (*qstate->env->detach_subs)(qstate);
3148 iq->num_target_queries = 0;
3149 iq->response = NULL;
3150 iq->fail_reply = NULL;
3151 verbose(VERB_ALGO, "cleared outbound list for next round");
3152 return next_state(iq, QUERYTARGETS_STATE);
3153 } else if(type == RESPONSE_TYPE_CNAME) {
3154 uint8_t* sname = NULL;
3155 size_t snamelen = 0;
3156 /* CNAME type responses get a query restart (i.e., get a
3157 * reset of the query state and go back to INIT_REQUEST_STATE).
3158 */
3159 verbose(VERB_DETAIL, "query response was CNAME");
3160 if(verbosity >= VERB_ALGO)
3161 log_dns_msg("cname msg", &iq->response->qinfo,
3162 iq->response->rep);
3163 /* if qtype is DS, check we have the right level of answer,
3164 * like grandchild answer but we need the middle, reject it */
3165 if(iq->qchase.qtype == LDNS_RR_TYPE_DS && !iq->dsns_point
3166 && !(iq->chase_flags&BIT_RD)
3167 && iter_ds_toolow(iq->response, iq->dp)
3168 && iter_dp_cangodown(&iq->qchase, iq->dp)) {
3169 outbound_list_clear(&iq->outlist);
3170 iq->num_current_queries = 0;
3171 fptr_ok(fptr_whitelist_modenv_detach_subs(
3172 qstate->env->detach_subs));
3173 (*qstate->env->detach_subs)(qstate);
3174 iq->num_target_queries = 0;
3175 return processDSNSFind(qstate, iq, id);
3176 }
3177 /* Process the CNAME response. */
3178 if(!handle_cname_response(qstate, iq, iq->response,
3179 &sname, &snamelen)) {
3180 errinf(qstate, "malloc failure, CNAME info");
3181 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3182 }
3183 /* cache the CNAME response under the current query */
3184 /* NOTE : set referral=1, so that rrsets get stored but not
3185 * the partial query answer (CNAME only). */
3186 /* prefetchleeway applied because this updates answer parts */
3187 if(!qstate->no_cache_store)
3188 iter_dns_store(qstate->env, &iq->response->qinfo,
3189 iq->response->rep, 1, qstate->prefetch_leeway,
3190 iq->dp&&iq->dp->has_parent_side_NS, NULL,
3191 qstate->query_flags, qstate->qstarttime);
3192 /* set the current request's qname to the new value. */
3193 iq->qchase.qname = sname;
3194 iq->qchase.qname_len = snamelen;
3195 if(qstate->env->auth_zones) {
3196 /* apply rpz qname triggers after cname */
3197 struct dns_msg* forged_response =
3198 rpz_callback_from_iterator_cname(qstate, iq);
3199 while(forged_response && reply_find_rrset_section_an(
3200 forged_response->rep, iq->qchase.qname,
3201 iq->qchase.qname_len, LDNS_RR_TYPE_CNAME,
3202 iq->qchase.qclass)) {
3203 /* another cname to follow */
3204 if(!handle_cname_response(qstate, iq, forged_response,
3205 &sname, &snamelen)) {
3206 errinf(qstate, "malloc failure, CNAME info");
3207 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3208 }
3209 iq->qchase.qname = sname;
3210 iq->qchase.qname_len = snamelen;
3211 forged_response =
3212 rpz_callback_from_iterator_cname(qstate, iq);
3213 }
3214 if(forged_response != NULL) {
3215 qstate->ext_state[id] = module_finished;
3216 qstate->return_rcode = LDNS_RCODE_NOERROR;
3217 qstate->return_msg = forged_response;
3218 iq->response = forged_response;
3219 next_state(iq, FINISHED_STATE);
3220 if(!iter_prepend(iq, qstate->return_msg, qstate->region)) {
3221 log_err("rpz: after cname, prepend rrsets: out of memory");
3222 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3223 }
3224 qstate->return_msg->qinfo = qstate->qinfo;
3225 return 0;
3226 }
3227 }
3228 /* Clear the query state, since this is a query restart. */
3229 iq->deleg_msg = NULL;
3230 iq->dp = NULL;
3231 iq->dsns_point = NULL;
3232 iq->auth_zone_response = 0;
3233 iq->sent_count = 0;
3234 iq->dp_target_count = 0;
3235 if(iq->minimisation_state != MINIMISE_STATE)
3236 /* Only count as query restart when it is not an extra
3237 * query as result of qname minimisation. */
3238 iq->query_restart_count++;
3239 if(qstate->env->cfg->qname_minimisation)
3240 iq->minimisation_state = INIT_MINIMISE_STATE;
3241
3242 /* stop current outstanding queries.
3243 * FIXME: should the outstanding queries be waited for and
3244 * handled? Say by a subquery that inherits the outbound_entry.
3245 */
3246 outbound_list_clear(&iq->outlist);
3247 iq->num_current_queries = 0;
3248 fptr_ok(fptr_whitelist_modenv_detach_subs(
3249 qstate->env->detach_subs));
3250 (*qstate->env->detach_subs)(qstate);
3251 iq->num_target_queries = 0;
3252 if(qstate->reply)
3253 sock_list_insert(&qstate->reply_origin,
3254 &qstate->reply->addr, qstate->reply->addrlen,
3255 qstate->region);
3256 verbose(VERB_ALGO, "cleared outbound list for query restart");
3257 /* go to INIT_REQUEST_STATE for new qname. */
3258 return next_state(iq, INIT_REQUEST_STATE);
3259 } else if(type == RESPONSE_TYPE_LAME) {
3260 /* Cache the LAMEness. */
3261 verbose(VERB_DETAIL, "query response was %sLAME",
3262 dnsseclame?"DNSSEC ":"");
3263 if(!dname_subdomain_c(iq->qchase.qname, iq->dp->name)) {
3264 log_err("mark lame: mismatch in qname and dpname");
3265 /* throwaway this reply below */
3266 } else if(qstate->reply) {
3267 /* need addr for lameness cache, but we may have
3268 * gotten this from cache, so test to be sure */
3269 if(!infra_set_lame(qstate->env->infra_cache,
3270 &qstate->reply->addr, qstate->reply->addrlen,
3271 iq->dp->name, iq->dp->namelen,
3272 *qstate->env->now, dnsseclame, 0,
3273 iq->qchase.qtype))
3274 log_err("mark host lame: out of memory");
3275 }
3276 } else if(type == RESPONSE_TYPE_REC_LAME) {
3277 /* Cache the LAMEness. */
3278 verbose(VERB_DETAIL, "query response REC_LAME: "
3279 "recursive but not authoritative server");
3280 if(!dname_subdomain_c(iq->qchase.qname, iq->dp->name)) {
3281 log_err("mark rec_lame: mismatch in qname and dpname");
3282 /* throwaway this reply below */
3283 } else if(qstate->reply) {
3284 /* need addr for lameness cache, but we may have
3285 * gotten this from cache, so test to be sure */
3286 verbose(VERB_DETAIL, "mark as REC_LAME");
3287 if(!infra_set_lame(qstate->env->infra_cache,
3288 &qstate->reply->addr, qstate->reply->addrlen,
3289 iq->dp->name, iq->dp->namelen,
3290 *qstate->env->now, 0, 1, iq->qchase.qtype))
3291 log_err("mark host lame: out of memory");
3292 }
3293 } else if(type == RESPONSE_TYPE_THROWAWAY) {
3294 /* LAME and THROWAWAY responses are handled the same way.
3295 * In this case, the event is just sent directly back to
3296 * the QUERYTARGETS_STATE without resetting anything,
3297 * because, clearly, the next target must be tried. */
3298 verbose(VERB_DETAIL, "query response was THROWAWAY");
3299 } else {
3300 log_warn("A query response came back with an unknown type: %d",
3301 (int)type);
3302 }
3303
3304 /* LAME, THROWAWAY and "unknown" all end up here.
3305 * Recycle to the QUERYTARGETS state to hopefully try a
3306 * different target. */
3307 if (qstate->env->cfg->qname_minimisation &&
3308 !qstate->env->cfg->qname_minimisation_strict)
3309 iq->minimisation_state = DONOT_MINIMISE_STATE;
3310 if(iq->auth_zone_response) {
3311 /* can we fallback? */
3312 iq->auth_zone_response = 0;
3313 if(!auth_zones_can_fallback(qstate->env->auth_zones,
3314 iq->dp->name, iq->dp->namelen, qstate->qinfo.qclass)) {
3315 verbose(VERB_ALGO, "auth zone response bad, and no"
3316 " fallback possible, servfail");
3317 errinf_dname(qstate, "response is bad, no fallback, "
3318 "for auth zone", iq->dp->name);
3319 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3320 }
3321 verbose(VERB_ALGO, "auth zone response was bad, "
3322 "fallback enabled");
3323 iq->auth_zone_avoid = 1;
3324 if(iq->dp->auth_dp) {
3325 /* we are using a dp for the auth zone, with no
3326 * nameservers, get one first */
3327 iq->dp = NULL;
3328 return next_state(iq, INIT_REQUEST_STATE);
3329 }
3330 }
3331 return next_state(iq, QUERYTARGETS_STATE);
3332 }
3333
3334 /**
3335 * Return priming query results to interested super querystates.
3336 *
3337 * Sets the delegation point and delegation message (not nonRD queries).
3338 * This is a callback from walk_supers.
3339 *
3340 * @param qstate: priming query state that finished.
3341 * @param id: module id.
3342 * @param forq: the qstate for which priming has been done.
3343 */
3344 static void
prime_supers(struct module_qstate * qstate,int id,struct module_qstate * forq)3345 prime_supers(struct module_qstate* qstate, int id, struct module_qstate* forq)
3346 {
3347 struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
3348 struct delegpt* dp = NULL;
3349
3350 log_assert(qstate->is_priming || foriq->wait_priming_stub);
3351 log_assert(qstate->return_rcode == LDNS_RCODE_NOERROR);
3352 /* Convert our response to a delegation point */
3353 dp = delegpt_from_message(qstate->return_msg, forq->region);
3354 if(!dp) {
3355 /* if there is no convertible delegation point, then
3356 * the ANSWER type was (presumably) a negative answer. */
3357 verbose(VERB_ALGO, "prime response was not a positive "
3358 "ANSWER; failing");
3359 foriq->dp = NULL;
3360 foriq->state = QUERYTARGETS_STATE;
3361 return;
3362 }
3363
3364 log_query_info(VERB_DETAIL, "priming successful for", &qstate->qinfo);
3365 delegpt_log(VERB_ALGO, dp);
3366 foriq->dp = dp;
3367 foriq->deleg_msg = dns_copy_msg(qstate->return_msg, forq->region);
3368 if(!foriq->deleg_msg) {
3369 log_err("copy prime response: out of memory");
3370 foriq->dp = NULL;
3371 foriq->state = QUERYTARGETS_STATE;
3372 return;
3373 }
3374
3375 /* root priming responses go to init stage 2, priming stub
3376 * responses to to stage 3. */
3377 if(foriq->wait_priming_stub) {
3378 foriq->state = INIT_REQUEST_3_STATE;
3379 foriq->wait_priming_stub = 0;
3380 } else foriq->state = INIT_REQUEST_2_STATE;
3381 /* because we are finished, the parent will be reactivated */
3382 }
3383
3384 /**
3385 * This handles the response to a priming query. This is used to handle both
3386 * root and stub priming responses. This is basically the equivalent of the
3387 * QUERY_RESP_STATE, but will not handle CNAME responses and will treat
3388 * REFERRALs as ANSWERS. It will also update and reactivate the originating
3389 * event.
3390 *
3391 * @param qstate: query state.
3392 * @param id: module id.
3393 * @return true if the event needs more immediate processing, false if not.
3394 * This state always returns false.
3395 */
3396 static int
processPrimeResponse(struct module_qstate * qstate,int id)3397 processPrimeResponse(struct module_qstate* qstate, int id)
3398 {
3399 struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
3400 enum response_type type;
3401 iq->response->rep->flags &= ~(BIT_RD|BIT_RA); /* ignore rec-lame */
3402 type = response_type_from_server(
3403 (int)((iq->chase_flags&BIT_RD) || iq->chase_to_rd),
3404 iq->response, &iq->qchase, iq->dp);
3405 if(type == RESPONSE_TYPE_ANSWER) {
3406 qstate->return_rcode = LDNS_RCODE_NOERROR;
3407 qstate->return_msg = iq->response;
3408 } else {
3409 errinf(qstate, "prime response did not get an answer");
3410 errinf_dname(qstate, "for", qstate->qinfo.qname);
3411 qstate->return_rcode = LDNS_RCODE_SERVFAIL;
3412 qstate->return_msg = NULL;
3413 }
3414
3415 /* validate the root or stub after priming (if enabled).
3416 * This is the same query as the prime query, but with validation.
3417 * Now that we are primed, the additional queries that validation
3418 * may need can be resolved. */
3419 if(qstate->env->cfg->harden_referral_path) {
3420 struct module_qstate* subq = NULL;
3421 log_nametypeclass(VERB_ALGO, "schedule prime validation",
3422 qstate->qinfo.qname, qstate->qinfo.qtype,
3423 qstate->qinfo.qclass);
3424 if(!generate_sub_request(qstate->qinfo.qname,
3425 qstate->qinfo.qname_len, qstate->qinfo.qtype,
3426 qstate->qinfo.qclass, qstate, id, iq,
3427 INIT_REQUEST_STATE, FINISHED_STATE, &subq, 1, 0)) {
3428 verbose(VERB_ALGO, "could not generate prime check");
3429 }
3430 generate_a_aaaa_check(qstate, iq, id);
3431 }
3432
3433 /* This event is finished. */
3434 qstate->ext_state[id] = module_finished;
3435 return 0;
3436 }
3437
3438 /**
3439 * Do final processing on responses to target queries. Events reach this
3440 * state after the iterative resolution algorithm terminates. This state is
3441 * responsible for reactivating the original event, and housekeeping related
3442 * to received target responses (caching, updating the current delegation
3443 * point, etc).
3444 * Callback from walk_supers for every super state that is interested in
3445 * the results from this query.
3446 *
3447 * @param qstate: query state.
3448 * @param id: module id.
3449 * @param forq: super query state.
3450 */
3451 static void
processTargetResponse(struct module_qstate * qstate,int id,struct module_qstate * forq)3452 processTargetResponse(struct module_qstate* qstate, int id,
3453 struct module_qstate* forq)
3454 {
3455 struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
3456 struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
3457 struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
3458 struct ub_packed_rrset_key* rrset;
3459 struct delegpt_ns* dpns;
3460 log_assert(qstate->return_rcode == LDNS_RCODE_NOERROR);
3461
3462 foriq->state = QUERYTARGETS_STATE;
3463 log_query_info(VERB_ALGO, "processTargetResponse", &qstate->qinfo);
3464 log_query_info(VERB_ALGO, "processTargetResponse super", &forq->qinfo);
3465
3466 /* Tell the originating event that this target query has finished
3467 * (regardless if it succeeded or not). */
3468 foriq->num_target_queries--;
3469
3470 /* check to see if parent event is still interested (in orig name). */
3471 if(!foriq->dp) {
3472 verbose(VERB_ALGO, "subq: parent not interested, was reset");
3473 return; /* not interested anymore */
3474 }
3475 dpns = delegpt_find_ns(foriq->dp, qstate->qinfo.qname,
3476 qstate->qinfo.qname_len);
3477 if(!dpns) {
3478 /* If not interested, just stop processing this event */
3479 verbose(VERB_ALGO, "subq: parent not interested anymore");
3480 /* could be because parent was jostled out of the cache,
3481 and a new identical query arrived, that does not want it*/
3482 return;
3483 }
3484
3485 /* if iq->query_for_pside_glue then add the pside_glue (marked lame) */
3486 if(iq->pside_glue) {
3487 /* if the pside_glue is NULL, then it could not be found,
3488 * the done_pside is already set when created and a cache
3489 * entry created in processFinished so nothing to do here */
3490 log_rrset_key(VERB_ALGO, "add parentside glue to dp",
3491 iq->pside_glue);
3492 if(!delegpt_add_rrset(foriq->dp, forq->region,
3493 iq->pside_glue, 1, NULL))
3494 log_err("out of memory adding pside glue");
3495 }
3496
3497 /* This response is relevant to the current query, so we
3498 * add (attempt to add, anyway) this target(s) and reactivate
3499 * the original event.
3500 * NOTE: we could only look for the AnswerRRset if the
3501 * response type was ANSWER. */
3502 rrset = reply_find_answer_rrset(&iq->qchase, qstate->return_msg->rep);
3503 if(rrset) {
3504 int additions = 0;
3505 /* if CNAMEs have been followed - add new NS to delegpt. */
3506 /* BTW. RFC 1918 says NS should not have got CNAMEs. Robust. */
3507 if(!delegpt_find_ns(foriq->dp, rrset->rk.dname,
3508 rrset->rk.dname_len)) {
3509 /* if dpns->lame then set newcname ns lame too */
3510 if(!delegpt_add_ns(foriq->dp, forq->region,
3511 rrset->rk.dname, dpns->lame, dpns->tls_auth_name,
3512 dpns->port))
3513 log_err("out of memory adding cnamed-ns");
3514 }
3515 /* if dpns->lame then set the address(es) lame too */
3516 if(!delegpt_add_rrset(foriq->dp, forq->region, rrset,
3517 dpns->lame, &additions))
3518 log_err("out of memory adding targets");
3519 if(!additions) {
3520 /* no new addresses, increase the nxns counter, like
3521 * this could be a list of wildcards with no new
3522 * addresses */
3523 target_count_increase_nx(foriq, 1);
3524 }
3525 verbose(VERB_ALGO, "added target response");
3526 delegpt_log(VERB_ALGO, foriq->dp);
3527 } else {
3528 verbose(VERB_ALGO, "iterator TargetResponse failed");
3529 delegpt_mark_neg(dpns, qstate->qinfo.qtype);
3530 dpns->resolved = 1; /* fail the target */
3531 if((dpns->got4 == 2 || !ie->supports_ipv4) &&
3532 (dpns->got6 == 2 || !ie->supports_ipv6) &&
3533 /* do not count cached answers */
3534 (qstate->reply_origin && qstate->reply_origin->len != 0)) {
3535 target_count_increase_nx(foriq, 1);
3536 }
3537 }
3538 }
3539
3540 /**
3541 * Process response for DS NS Find queries, that attempt to find the delegation
3542 * point where we ask the DS query from.
3543 *
3544 * @param qstate: query state.
3545 * @param id: module id.
3546 * @param forq: super query state.
3547 */
3548 static void
processDSNSResponse(struct module_qstate * qstate,int id,struct module_qstate * forq)3549 processDSNSResponse(struct module_qstate* qstate, int id,
3550 struct module_qstate* forq)
3551 {
3552 struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
3553
3554 /* if the finished (iq->response) query has no NS set: continue
3555 * up to look for the right dp; nothing to change, do DPNSstate */
3556 if(qstate->return_rcode != LDNS_RCODE_NOERROR)
3557 return; /* seek further */
3558 /* find the NS RRset (without allowing CNAMEs) */
3559 if(!reply_find_rrset(qstate->return_msg->rep, qstate->qinfo.qname,
3560 qstate->qinfo.qname_len, LDNS_RR_TYPE_NS,
3561 qstate->qinfo.qclass)){
3562 return; /* seek further */
3563 }
3564
3565 /* else, store as DP and continue at querytargets */
3566 foriq->state = QUERYTARGETS_STATE;
3567 foriq->dp = delegpt_from_message(qstate->return_msg, forq->region);
3568 if(!foriq->dp) {
3569 log_err("out of memory in dsns dp alloc");
3570 errinf(qstate, "malloc failure, in DS search");
3571 return; /* dp==NULL in QUERYTARGETS makes SERVFAIL */
3572 }
3573 /* success, go query the querytargets in the new dp (and go down) */
3574 }
3575
3576 /**
3577 * Process response for qclass=ANY queries for a particular class.
3578 * Append to result or error-exit.
3579 *
3580 * @param qstate: query state.
3581 * @param id: module id.
3582 * @param forq: super query state.
3583 */
3584 static void
processClassResponse(struct module_qstate * qstate,int id,struct module_qstate * forq)3585 processClassResponse(struct module_qstate* qstate, int id,
3586 struct module_qstate* forq)
3587 {
3588 struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
3589 struct dns_msg* from = qstate->return_msg;
3590 log_query_info(VERB_ALGO, "processClassResponse", &qstate->qinfo);
3591 log_query_info(VERB_ALGO, "processClassResponse super", &forq->qinfo);
3592 if(qstate->return_rcode != LDNS_RCODE_NOERROR) {
3593 /* cause servfail for qclass ANY query */
3594 foriq->response = NULL;
3595 foriq->state = FINISHED_STATE;
3596 return;
3597 }
3598 /* append result */
3599 if(!foriq->response) {
3600 /* allocate the response: copy RCODE, sec_state */
3601 foriq->response = dns_copy_msg(from, forq->region);
3602 if(!foriq->response) {
3603 log_err("malloc failed for qclass ANY response");
3604 foriq->state = FINISHED_STATE;
3605 return;
3606 }
3607 foriq->response->qinfo.qclass = forq->qinfo.qclass;
3608 /* qclass ANY does not receive the AA flag on replies */
3609 foriq->response->rep->authoritative = 0;
3610 } else {
3611 struct dns_msg* to = foriq->response;
3612 /* add _from_ this response _to_ existing collection */
3613 /* if there are records, copy RCODE */
3614 /* lower sec_state if this message is lower */
3615 if(from->rep->rrset_count != 0) {
3616 size_t n = from->rep->rrset_count+to->rep->rrset_count;
3617 struct ub_packed_rrset_key** dest, **d;
3618 /* copy appropriate rcode */
3619 to->rep->flags = from->rep->flags;
3620 /* copy rrsets */
3621 if(from->rep->rrset_count > RR_COUNT_MAX ||
3622 to->rep->rrset_count > RR_COUNT_MAX) {
3623 log_err("malloc failed (too many rrsets) in collect ANY");
3624 foriq->state = FINISHED_STATE;
3625 return; /* integer overflow protection */
3626 }
3627 dest = regional_alloc(forq->region, sizeof(dest[0])*n);
3628 if(!dest) {
3629 log_err("malloc failed in collect ANY");
3630 foriq->state = FINISHED_STATE;
3631 return;
3632 }
3633 d = dest;
3634 /* copy AN */
3635 memcpy(dest, to->rep->rrsets, to->rep->an_numrrsets
3636 * sizeof(dest[0]));
3637 dest += to->rep->an_numrrsets;
3638 memcpy(dest, from->rep->rrsets, from->rep->an_numrrsets
3639 * sizeof(dest[0]));
3640 dest += from->rep->an_numrrsets;
3641 /* copy NS */
3642 memcpy(dest, to->rep->rrsets+to->rep->an_numrrsets,
3643 to->rep->ns_numrrsets * sizeof(dest[0]));
3644 dest += to->rep->ns_numrrsets;
3645 memcpy(dest, from->rep->rrsets+from->rep->an_numrrsets,
3646 from->rep->ns_numrrsets * sizeof(dest[0]));
3647 dest += from->rep->ns_numrrsets;
3648 /* copy AR */
3649 memcpy(dest, to->rep->rrsets+to->rep->an_numrrsets+
3650 to->rep->ns_numrrsets,
3651 to->rep->ar_numrrsets * sizeof(dest[0]));
3652 dest += to->rep->ar_numrrsets;
3653 memcpy(dest, from->rep->rrsets+from->rep->an_numrrsets+
3654 from->rep->ns_numrrsets,
3655 from->rep->ar_numrrsets * sizeof(dest[0]));
3656 /* update counts */
3657 to->rep->rrsets = d;
3658 to->rep->an_numrrsets += from->rep->an_numrrsets;
3659 to->rep->ns_numrrsets += from->rep->ns_numrrsets;
3660 to->rep->ar_numrrsets += from->rep->ar_numrrsets;
3661 to->rep->rrset_count = n;
3662 }
3663 if(from->rep->security < to->rep->security) /* lowest sec */
3664 to->rep->security = from->rep->security;
3665 if(from->rep->qdcount != 0) /* insert qd if appropriate */
3666 to->rep->qdcount = from->rep->qdcount;
3667 if(from->rep->ttl < to->rep->ttl) /* use smallest TTL */
3668 to->rep->ttl = from->rep->ttl;
3669 if(from->rep->prefetch_ttl < to->rep->prefetch_ttl)
3670 to->rep->prefetch_ttl = from->rep->prefetch_ttl;
3671 if(from->rep->serve_expired_ttl < to->rep->serve_expired_ttl)
3672 to->rep->serve_expired_ttl = from->rep->serve_expired_ttl;
3673 }
3674 /* are we done? */
3675 foriq->num_current_queries --;
3676 if(foriq->num_current_queries == 0)
3677 foriq->state = FINISHED_STATE;
3678 }
3679
3680 /**
3681 * Collect class ANY responses and make them into one response. This
3682 * state is started and it creates queries for all classes (that have
3683 * root hints). The answers are then collected.
3684 *
3685 * @param qstate: query state.
3686 * @param id: module id.
3687 * @return true if the event needs more immediate processing, false if not.
3688 */
3689 static int
processCollectClass(struct module_qstate * qstate,int id)3690 processCollectClass(struct module_qstate* qstate, int id)
3691 {
3692 struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
3693 struct module_qstate* subq;
3694 /* If qchase.qclass == 0 then send out queries for all classes.
3695 * Otherwise, do nothing (wait for all answers to arrive and the
3696 * processClassResponse to put them together, and that moves us
3697 * towards the Finished state when done. */
3698 if(iq->qchase.qclass == 0) {
3699 uint16_t c = 0;
3700 iq->qchase.qclass = LDNS_RR_CLASS_ANY;
3701 while(iter_get_next_root(qstate->env->hints,
3702 qstate->env->fwds, &c)) {
3703 /* generate query for this class */
3704 log_nametypeclass(VERB_ALGO, "spawn collect query",
3705 qstate->qinfo.qname, qstate->qinfo.qtype, c);
3706 if(!generate_sub_request(qstate->qinfo.qname,
3707 qstate->qinfo.qname_len, qstate->qinfo.qtype,
3708 c, qstate, id, iq, INIT_REQUEST_STATE,
3709 FINISHED_STATE, &subq,
3710 (int)!(qstate->query_flags&BIT_CD), 0)) {
3711 errinf(qstate, "could not generate class ANY"
3712 " lookup query");
3713 return error_response(qstate, id,
3714 LDNS_RCODE_SERVFAIL);
3715 }
3716 /* ignore subq, no special init required */
3717 iq->num_current_queries ++;
3718 if(c == 0xffff)
3719 break;
3720 else c++;
3721 }
3722 /* if no roots are configured at all, return */
3723 if(iq->num_current_queries == 0) {
3724 verbose(VERB_ALGO, "No root hints or fwds, giving up "
3725 "on qclass ANY");
3726 return error_response(qstate, id, LDNS_RCODE_REFUSED);
3727 }
3728 /* return false, wait for queries to return */
3729 }
3730 /* if woke up here because of an answer, wait for more answers */
3731 return 0;
3732 }
3733
3734 /**
3735 * This handles the final state for first-tier responses (i.e., responses to
3736 * externally generated queries).
3737 *
3738 * @param qstate: query state.
3739 * @param iq: iterator query state.
3740 * @param id: module id.
3741 * @return true if the event needs more processing, false if not. Since this
3742 * is the final state for an event, it always returns false.
3743 */
3744 static int
processFinished(struct module_qstate * qstate,struct iter_qstate * iq,int id)3745 processFinished(struct module_qstate* qstate, struct iter_qstate* iq,
3746 int id)
3747 {
3748 log_query_info(VERB_QUERY, "finishing processing for",
3749 &qstate->qinfo);
3750
3751 /* store negative cache element for parent side glue. */
3752 if(!qstate->no_cache_store && iq->query_for_pside_glue
3753 && !iq->pside_glue)
3754 iter_store_parentside_neg(qstate->env, &qstate->qinfo,
3755 iq->deleg_msg?iq->deleg_msg->rep:
3756 (iq->response?iq->response->rep:NULL));
3757 if(!iq->response) {
3758 verbose(VERB_ALGO, "No response is set, servfail");
3759 errinf(qstate, "(no response found at query finish)");
3760 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3761 }
3762
3763 /* Make sure that the RA flag is set (since the presence of
3764 * this module means that recursion is available) */
3765 iq->response->rep->flags |= BIT_RA;
3766
3767 /* Clear the AA flag */
3768 /* FIXME: does this action go here or in some other module? */
3769 iq->response->rep->flags &= ~BIT_AA;
3770
3771 /* make sure QR flag is on */
3772 iq->response->rep->flags |= BIT_QR;
3773
3774 /* we have finished processing this query */
3775 qstate->ext_state[id] = module_finished;
3776
3777 /* TODO: we are using a private TTL, trim the response. */
3778 /* if (mPrivateTTL > 0){IterUtils.setPrivateTTL(resp, mPrivateTTL); } */
3779
3780 /* prepend any items we have accumulated */
3781 if(iq->an_prepend_list || iq->ns_prepend_list) {
3782 if(!iter_prepend(iq, iq->response, qstate->region)) {
3783 log_err("prepend rrsets: out of memory");
3784 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3785 }
3786 /* reset the query name back */
3787 iq->response->qinfo = qstate->qinfo;
3788 /* the security state depends on the combination */
3789 iq->response->rep->security = sec_status_unchecked;
3790 /* store message with the finished prepended items,
3791 * but only if we did recursion. The nonrecursion referral
3792 * from cache does not need to be stored in the msg cache. */
3793 if(!qstate->no_cache_store && qstate->query_flags&BIT_RD) {
3794 iter_dns_store(qstate->env, &qstate->qinfo,
3795 iq->response->rep, 0, qstate->prefetch_leeway,
3796 iq->dp&&iq->dp->has_parent_side_NS,
3797 qstate->region, qstate->query_flags,
3798 qstate->qstarttime);
3799 }
3800 }
3801 qstate->return_rcode = LDNS_RCODE_NOERROR;
3802 qstate->return_msg = iq->response;
3803 return 0;
3804 }
3805
3806 /*
3807 * Return priming query results to interested super querystates.
3808 *
3809 * Sets the delegation point and delegation message (not nonRD queries).
3810 * This is a callback from walk_supers.
3811 *
3812 * @param qstate: query state that finished.
3813 * @param id: module id.
3814 * @param super: the qstate to inform.
3815 */
3816 void
iter_inform_super(struct module_qstate * qstate,int id,struct module_qstate * super)3817 iter_inform_super(struct module_qstate* qstate, int id,
3818 struct module_qstate* super)
3819 {
3820 if(!qstate->is_priming && super->qinfo.qclass == LDNS_RR_CLASS_ANY)
3821 processClassResponse(qstate, id, super);
3822 else if(super->qinfo.qtype == LDNS_RR_TYPE_DS && ((struct iter_qstate*)
3823 super->minfo[id])->state == DSNS_FIND_STATE)
3824 processDSNSResponse(qstate, id, super);
3825 else if(qstate->return_rcode != LDNS_RCODE_NOERROR)
3826 error_supers(qstate, id, super);
3827 else if(qstate->is_priming)
3828 prime_supers(qstate, id, super);
3829 else processTargetResponse(qstate, id, super);
3830 }
3831
3832 /**
3833 * Handle iterator state.
3834 * Handle events. This is the real processing loop for events, responsible
3835 * for moving events through the various states. If a processing method
3836 * returns true, then it will be advanced to the next state. If false, then
3837 * processing will stop.
3838 *
3839 * @param qstate: query state.
3840 * @param ie: iterator shared global environment.
3841 * @param iq: iterator query state.
3842 * @param id: module id.
3843 */
3844 static void
iter_handle(struct module_qstate * qstate,struct iter_qstate * iq,struct iter_env * ie,int id)3845 iter_handle(struct module_qstate* qstate, struct iter_qstate* iq,
3846 struct iter_env* ie, int id)
3847 {
3848 int cont = 1;
3849 while(cont) {
3850 verbose(VERB_ALGO, "iter_handle processing q with state %s",
3851 iter_state_to_string(iq->state));
3852 switch(iq->state) {
3853 case INIT_REQUEST_STATE:
3854 cont = processInitRequest(qstate, iq, ie, id);
3855 break;
3856 case INIT_REQUEST_2_STATE:
3857 cont = processInitRequest2(qstate, iq, id);
3858 break;
3859 case INIT_REQUEST_3_STATE:
3860 cont = processInitRequest3(qstate, iq, id);
3861 break;
3862 case QUERYTARGETS_STATE:
3863 cont = processQueryTargets(qstate, iq, ie, id);
3864 break;
3865 case QUERY_RESP_STATE:
3866 cont = processQueryResponse(qstate, iq, ie, id);
3867 break;
3868 case PRIME_RESP_STATE:
3869 cont = processPrimeResponse(qstate, id);
3870 break;
3871 case COLLECT_CLASS_STATE:
3872 cont = processCollectClass(qstate, id);
3873 break;
3874 case DSNS_FIND_STATE:
3875 cont = processDSNSFind(qstate, iq, id);
3876 break;
3877 case FINISHED_STATE:
3878 cont = processFinished(qstate, iq, id);
3879 break;
3880 default:
3881 log_warn("iterator: invalid state: %d",
3882 iq->state);
3883 cont = 0;
3884 break;
3885 }
3886 }
3887 }
3888
3889 /**
3890 * This is the primary entry point for processing request events. Note that
3891 * this method should only be used by external modules.
3892 * @param qstate: query state.
3893 * @param ie: iterator shared global environment.
3894 * @param iq: iterator query state.
3895 * @param id: module id.
3896 */
3897 static void
process_request(struct module_qstate * qstate,struct iter_qstate * iq,struct iter_env * ie,int id)3898 process_request(struct module_qstate* qstate, struct iter_qstate* iq,
3899 struct iter_env* ie, int id)
3900 {
3901 /* external requests start in the INIT state, and finish using the
3902 * FINISHED state. */
3903 iq->state = INIT_REQUEST_STATE;
3904 iq->final_state = FINISHED_STATE;
3905 verbose(VERB_ALGO, "process_request: new external request event");
3906 iter_handle(qstate, iq, ie, id);
3907 }
3908
3909 /** process authoritative server reply */
3910 static void
process_response(struct module_qstate * qstate,struct iter_qstate * iq,struct iter_env * ie,int id,struct outbound_entry * outbound,enum module_ev event)3911 process_response(struct module_qstate* qstate, struct iter_qstate* iq,
3912 struct iter_env* ie, int id, struct outbound_entry* outbound,
3913 enum module_ev event)
3914 {
3915 struct msg_parse* prs;
3916 struct edns_data edns;
3917 sldns_buffer* pkt;
3918
3919 verbose(VERB_ALGO, "process_response: new external response event");
3920 iq->response = NULL;
3921 iq->state = QUERY_RESP_STATE;
3922 if(event == module_event_noreply || event == module_event_error) {
3923 if(event == module_event_noreply && iq->timeout_count >= 3 &&
3924 qstate->env->cfg->use_caps_bits_for_id &&
3925 !iq->caps_fallback && !is_caps_whitelisted(ie, iq)) {
3926 /* start fallback */
3927 iq->caps_fallback = 1;
3928 iq->caps_server = 0;
3929 iq->caps_reply = NULL;
3930 iq->caps_response = NULL;
3931 iq->caps_minimisation_state = DONOT_MINIMISE_STATE;
3932 iq->state = QUERYTARGETS_STATE;
3933 iq->num_current_queries--;
3934 /* need fresh attempts for the 0x20 fallback, if
3935 * that was the cause for the failure */
3936 iter_dec_attempts(iq->dp, 3, ie->outbound_msg_retry);
3937 verbose(VERB_DETAIL, "Capsforid: timeouts, starting fallback");
3938 goto handle_it;
3939 }
3940 goto handle_it;
3941 }
3942 if( (event != module_event_reply && event != module_event_capsfail)
3943 || !qstate->reply) {
3944 log_err("Bad event combined with response");
3945 outbound_list_remove(&iq->outlist, outbound);
3946 errinf(qstate, "module iterator received wrong internal event with a response message");
3947 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3948 return;
3949 }
3950
3951 /* parse message */
3952 iq->fail_reply = qstate->reply;
3953 prs = (struct msg_parse*)regional_alloc(qstate->env->scratch,
3954 sizeof(struct msg_parse));
3955 if(!prs) {
3956 log_err("out of memory on incoming message");
3957 /* like packet got dropped */
3958 goto handle_it;
3959 }
3960 memset(prs, 0, sizeof(*prs));
3961 memset(&edns, 0, sizeof(edns));
3962 pkt = qstate->reply->c->buffer;
3963 sldns_buffer_set_position(pkt, 0);
3964 if(parse_packet(pkt, prs, qstate->env->scratch) != LDNS_RCODE_NOERROR) {
3965 verbose(VERB_ALGO, "parse error on reply packet");
3966 iq->parse_failures++;
3967 goto handle_it;
3968 }
3969 /* edns is not examined, but removed from message to help cache */
3970 if(parse_extract_edns_from_response_msg(prs, &edns, qstate->env->scratch) !=
3971 LDNS_RCODE_NOERROR) {
3972 iq->parse_failures++;
3973 goto handle_it;
3974 }
3975
3976 /* Copy the edns options we may got from the back end */
3977 if(edns.opt_list_in) {
3978 qstate->edns_opts_back_in = edns_opt_copy_region(edns.opt_list_in,
3979 qstate->region);
3980 if(!qstate->edns_opts_back_in) {
3981 log_err("out of memory on incoming message");
3982 /* like packet got dropped */
3983 goto handle_it;
3984 }
3985 if(!inplace_cb_edns_back_parsed_call(qstate->env, qstate)) {
3986 log_err("unable to call edns_back_parsed callback");
3987 goto handle_it;
3988 }
3989 }
3990
3991 /* remove CD-bit, we asked for in case we handle validation ourself */
3992 prs->flags &= ~BIT_CD;
3993
3994 /* normalize and sanitize: easy to delete items from linked lists */
3995 if(!scrub_message(pkt, prs, &iq->qinfo_out, iq->dp->name,
3996 qstate->env->scratch, qstate->env, ie)) {
3997 /* if 0x20 enabled, start fallback, but we have no message */
3998 if(event == module_event_capsfail && !iq->caps_fallback) {
3999 iq->caps_fallback = 1;
4000 iq->caps_server = 0;
4001 iq->caps_reply = NULL;
4002 iq->caps_response = NULL;
4003 iq->caps_minimisation_state = DONOT_MINIMISE_STATE;
4004 iq->state = QUERYTARGETS_STATE;
4005 iq->num_current_queries--;
4006 verbose(VERB_DETAIL, "Capsforid: scrub failed, starting fallback with no response");
4007 }
4008 iq->scrub_failures++;
4009 goto handle_it;
4010 }
4011
4012 /* allocate response dns_msg in region */
4013 iq->response = dns_alloc_msg(pkt, prs, qstate->region);
4014 if(!iq->response)
4015 goto handle_it;
4016 log_query_info(VERB_DETAIL, "response for", &qstate->qinfo);
4017 log_name_addr(VERB_DETAIL, "reply from", iq->dp->name,
4018 &qstate->reply->addr, qstate->reply->addrlen);
4019 if(verbosity >= VERB_ALGO)
4020 log_dns_msg("incoming scrubbed packet:", &iq->response->qinfo,
4021 iq->response->rep);
4022
4023 if(event == module_event_capsfail || iq->caps_fallback) {
4024 if(qstate->env->cfg->qname_minimisation &&
4025 iq->minimisation_state != DONOT_MINIMISE_STATE) {
4026 /* Skip QNAME minimisation for next query, since that
4027 * one has to match the current query. */
4028 iq->minimisation_state = SKIP_MINIMISE_STATE;
4029 }
4030 /* for fallback we care about main answer, not additionals */
4031 /* removing that makes comparison more likely to succeed */
4032 caps_strip_reply(iq->response->rep);
4033
4034 if(iq->caps_fallback &&
4035 iq->caps_minimisation_state != iq->minimisation_state) {
4036 /* QNAME minimisation state has changed, restart caps
4037 * fallback. */
4038 iq->caps_fallback = 0;
4039 }
4040
4041 if(!iq->caps_fallback) {
4042 /* start fallback */
4043 iq->caps_fallback = 1;
4044 iq->caps_server = 0;
4045 iq->caps_reply = iq->response->rep;
4046 iq->caps_response = iq->response;
4047 iq->caps_minimisation_state = iq->minimisation_state;
4048 iq->state = QUERYTARGETS_STATE;
4049 iq->num_current_queries--;
4050 verbose(VERB_DETAIL, "Capsforid: starting fallback");
4051 goto handle_it;
4052 } else {
4053 /* check if reply is the same, otherwise, fail */
4054 if(!iq->caps_reply) {
4055 iq->caps_reply = iq->response->rep;
4056 iq->caps_response = iq->response;
4057 iq->caps_server = -1; /*become zero at ++,
4058 so that we start the full set of trials */
4059 } else if(caps_failed_rcode(iq->caps_reply) &&
4060 !caps_failed_rcode(iq->response->rep)) {
4061 /* prefer to upgrade to non-SERVFAIL */
4062 iq->caps_reply = iq->response->rep;
4063 iq->caps_response = iq->response;
4064 } else if(!caps_failed_rcode(iq->caps_reply) &&
4065 caps_failed_rcode(iq->response->rep)) {
4066 /* if we have non-SERVFAIL as answer then
4067 * we can ignore SERVFAILs for the equality
4068 * comparison */
4069 /* no instructions here, skip other else */
4070 } else if(caps_failed_rcode(iq->caps_reply) &&
4071 caps_failed_rcode(iq->response->rep)) {
4072 /* failure is same as other failure in fallbk*/
4073 /* no instructions here, skip other else */
4074 } else if(!reply_equal(iq->response->rep, iq->caps_reply,
4075 qstate->env->scratch)) {
4076 verbose(VERB_DETAIL, "Capsforid fallback: "
4077 "getting different replies, failed");
4078 outbound_list_remove(&iq->outlist, outbound);
4079 errinf(qstate, "0x20 failed, then got different replies in fallback");
4080 (void)error_response(qstate, id,
4081 LDNS_RCODE_SERVFAIL);
4082 return;
4083 }
4084 /* continue the fallback procedure at next server */
4085 iq->caps_server++;
4086 iq->state = QUERYTARGETS_STATE;
4087 iq->num_current_queries--;
4088 verbose(VERB_DETAIL, "Capsforid: reply is equal. "
4089 "go to next fallback");
4090 goto handle_it;
4091 }
4092 }
4093 iq->caps_fallback = 0; /* if we were in fallback, 0x20 is OK now */
4094
4095 handle_it:
4096 outbound_list_remove(&iq->outlist, outbound);
4097 iter_handle(qstate, iq, ie, id);
4098 }
4099
4100 void
iter_operate(struct module_qstate * qstate,enum module_ev event,int id,struct outbound_entry * outbound)4101 iter_operate(struct module_qstate* qstate, enum module_ev event, int id,
4102 struct outbound_entry* outbound)
4103 {
4104 struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
4105 struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
4106 verbose(VERB_QUERY, "iterator[module %d] operate: extstate:%s event:%s",
4107 id, strextstate(qstate->ext_state[id]), strmodulevent(event));
4108 if(iq) log_query_info(VERB_QUERY, "iterator operate: query",
4109 &qstate->qinfo);
4110 if(iq && qstate->qinfo.qname != iq->qchase.qname)
4111 log_query_info(VERB_QUERY, "iterator operate: chased to",
4112 &iq->qchase);
4113
4114 /* perform iterator state machine */
4115 if((event == module_event_new || event == module_event_pass) &&
4116 iq == NULL) {
4117 if(!iter_new(qstate, id)) {
4118 errinf(qstate, "malloc failure, new iterator module allocation");
4119 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
4120 return;
4121 }
4122 iq = (struct iter_qstate*)qstate->minfo[id];
4123 process_request(qstate, iq, ie, id);
4124 return;
4125 }
4126 if(iq && event == module_event_pass) {
4127 iter_handle(qstate, iq, ie, id);
4128 return;
4129 }
4130 if(iq && outbound) {
4131 process_response(qstate, iq, ie, id, outbound, event);
4132 return;
4133 }
4134 if(event == module_event_error) {
4135 verbose(VERB_ALGO, "got called with event error, giving up");
4136 errinf(qstate, "iterator module got the error event");
4137 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
4138 return;
4139 }
4140
4141 log_err("bad event for iterator");
4142 errinf(qstate, "iterator module received wrong event");
4143 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
4144 }
4145
4146 void
iter_clear(struct module_qstate * qstate,int id)4147 iter_clear(struct module_qstate* qstate, int id)
4148 {
4149 struct iter_qstate* iq;
4150 if(!qstate)
4151 return;
4152 iq = (struct iter_qstate*)qstate->minfo[id];
4153 if(iq) {
4154 outbound_list_clear(&iq->outlist);
4155 if(iq->target_count && --iq->target_count[TARGET_COUNT_REF] == 0) {
4156 free(iq->target_count);
4157 if(*iq->nxns_dp) free(*iq->nxns_dp);
4158 free(iq->nxns_dp);
4159 }
4160 iq->num_current_queries = 0;
4161 }
4162 qstate->minfo[id] = NULL;
4163 }
4164
4165 size_t
iter_get_mem(struct module_env * env,int id)4166 iter_get_mem(struct module_env* env, int id)
4167 {
4168 struct iter_env* ie = (struct iter_env*)env->modinfo[id];
4169 if(!ie)
4170 return 0;
4171 return sizeof(*ie) + sizeof(int)*((size_t)ie->max_dependency_depth+1)
4172 + donotq_get_mem(ie->donotq) + priv_get_mem(ie->priv);
4173 }
4174
4175 /**
4176 * The iterator function block
4177 */
4178 static struct module_func_block iter_block = {
4179 "iterator",
4180 &iter_init, &iter_deinit, &iter_operate, &iter_inform_super,
4181 &iter_clear, &iter_get_mem
4182 };
4183
4184 struct module_func_block*
iter_get_funcblock(void)4185 iter_get_funcblock(void)
4186 {
4187 return &iter_block;
4188 }
4189
4190 const char*
iter_state_to_string(enum iter_state state)4191 iter_state_to_string(enum iter_state state)
4192 {
4193 switch (state)
4194 {
4195 case INIT_REQUEST_STATE :
4196 return "INIT REQUEST STATE";
4197 case INIT_REQUEST_2_STATE :
4198 return "INIT REQUEST STATE (stage 2)";
4199 case INIT_REQUEST_3_STATE:
4200 return "INIT REQUEST STATE (stage 3)";
4201 case QUERYTARGETS_STATE :
4202 return "QUERY TARGETS STATE";
4203 case PRIME_RESP_STATE :
4204 return "PRIME RESPONSE STATE";
4205 case COLLECT_CLASS_STATE :
4206 return "COLLECT CLASS STATE";
4207 case DSNS_FIND_STATE :
4208 return "DSNS FIND STATE";
4209 case QUERY_RESP_STATE :
4210 return "QUERY RESPONSE STATE";
4211 case FINISHED_STATE :
4212 return "FINISHED RESPONSE STATE";
4213 default :
4214 return "UNKNOWN ITER STATE";
4215 }
4216 }
4217
4218 int
iter_state_is_responsestate(enum iter_state s)4219 iter_state_is_responsestate(enum iter_state s)
4220 {
4221 switch(s) {
4222 case INIT_REQUEST_STATE :
4223 case INIT_REQUEST_2_STATE :
4224 case INIT_REQUEST_3_STATE :
4225 case QUERYTARGETS_STATE :
4226 case COLLECT_CLASS_STATE :
4227 return 0;
4228 default:
4229 break;
4230 }
4231 return 1;
4232 }
4233